MOBOTIX MX M73A-LSA-WG, MX M73A-RJ45-WG User guide

Download

User Manual

AI-TECHVideo Analytics Apps

©2019MOBOTIX AG

TOC

Support 5

Imprint 7

Legal information 8

AI-Dashboard embedded for data management 11

AI-Dash - configuration overview 12

AI-Dash - Administrator parameters 17

Event notification 30

Custom server compliant event notification 51

AI-Dash - troubleshooting 56

AI-SECURITY 59

AI-SECURITY - camera positions 59

AI-INTRUSION 60

AI-LOITERING 71

AI-LOST 75

AI-RETAIL3 81

AI-RETAIL - camera positions 81

Recomended distances 82

AI-PEOPLE 82

AI-PEOPLE: Aggregate counting 88

AI-CROWD 90

AI-OVERCROWD 95

AI-TRAFFIC 99

AI-TRAFFIC - camera positions 103

AI-ROAD 3D 104

AI-ROAD 3D: 3D calibration – Step 1 108

AI-ROAD 3D: 3D calibration – Step 2 - Case A 109

AI-ROAD 3D: 3D calibration – Step 2 - Case B 110

2 / 153

AI-ROAD 3D: 3D calibration – Step 3 - Case A 111

AI-ROAD 3D: 3D calibration – Step 3 - Case B 112

AI-ROAD 3D: 3D calibration examples 113

AI-ROAD 3D: 3D - troubleshooting 113

AI-INCIDENT 115

AI-PARKING 119

AI-PARKING - camera positions 119

AI-PARKING 120

AI-BIO 123

AI-BIO - camera positions 123

Recommended distances 124

AI-BIO 125

AI-OCCUPANCY 127

AI-OCCUPANCY - camera positions 127

AI-OCCUPANCY 128

AI-OVEROCCUPANCY 131

AI-OVEROCCUPANCY - camera positions 131

AI-OVEROCCUPANCY 132

AI-HEAT 135

AI-HEAT 136

AI-SPILL 139

AI-SMOKE 143

AI-FIRE 147

List of figures 151

3 / 153

Support

If you need technical support, please contact your MOBOTIX-dealer. If your questions cannot be answered immediately, your reseller will forward your inquiries through the appropriate channels to ensure a quick response.

If you have Internet access, you can download additional documentation and software updates from the MOBOTIX-helpdesk. Please visit:

www.mobotix.com

5 / 153

Imprint

This document is part of products by MOBOTIX AG, hereinafter referred to as the manufacturer, and describes the use and configuration of AI-TECHApps on the camera and its components.

Changes, errors and misprints excepted.

This document is protected by copyright. The information contained herein may not be passed on to third parties without the express permission of the manufacturer. Contraventions will result in claims for damages.

Patent and copy protection

In the event that a patent, utility model or design protection is granted, all rights are reserved. Trademarks and product names are trademarks or registered trademarks of their respective companies or organizations.

Adress

MOBOTIX AG

Kaiserstrasse

67722 Langmeil

Germany

Tel.: +49 6302 9816-0

E-Mail: info@mobotix.com

Internet: www.mobotix.com

Support

See Support, p. 5.

7 / 153

Legal information

special export regulations

Cameras with thermal imaging sensors ("thermal imaging cameras") are subject to the special sanctions and export regulations of the USA, including the ITAR (International Traffic in Arms Regulation):

n According to the current US sanctions and export regulations, cameras with thermal image sensors or

parts thereof may not be delivered to countries or regions where the US has imposed an embargo,

unless a special exemption has been granted. At present, this applies in particular to the states/regions

of Syria, Iran, Cuba, North Korea, Sudan and Crimea. Furthermore, the corresponding ban on delivery

also applies to all persons and institutions listed in "The Denied Persons List" (see www.bis.doc.gov >

Policy Guidance > Lists of Parties of Concern; https://www.treasury.gov/resource-center-

/sanctions/sdn-list/pages/default.aspx).

n These cameras and the thermal imaging sensors used therein shall not be used in or be used in the

design, development or production of nuclear, biological or chemical weapons.

Legal aspects of video and audio recording:

When using MOBOTIX AG products, the data protection regulations for video and audio recording must be observed. epending on the state law and the place of installation, the recording of video and audio data may be subject to special requirements or prohibited. All users of MOBOTIX-Products are therefore requested to inform themselves about the currently valid regulations and to comply with them. MOBOTIX AG assumes no responsibility for product use that does not conform to legality.

Declaration of conformity.

The products of MOBOTIX AG are certified according to the applicable directives of the EU and other countries. You can find MOBOTIX AG's declarations of conformity for its products on the following website: www.-

mobotix.com under Support> Download Center> Documentation> Certificate & Declaration of Conformity.

Declaration of RoHS

The products of MOBOTIX AG comply with the requirements resulting from §5 ElektroG and the RoHS Directive 2011/65/EU, as far as they fall within the scope of these regulations. (die RoHS-declaration of MOBOTIX are available on the following website: www.mobotix.com under Support>Download Center> Documentation> Broschures & Instructions > Certificates).

Disposal

Electrical and electronic products contain many recyclable materials. Therefore, dispose MOBOTIX at the end of their service life in accordance with the applicable legal provisions and regulations (e.g. return to a municipal collection point). -Products must not be disposed of with household waste! Dispose of any battery in the product separately from the product (the respective product manuals contain a corresponding note if the product contains a battery).

8 / 153

Legal information

Disclaimer

MOBOTIX AG is not liable for damage caused by improper handling of its products, failure to observe the operating instructions and the relevant regulations. Our General Terms and Conditions apply, the current version of which can be found to download at www.mobotix.com .

9 / 153

AI-Dashboard embedded for data

management

The data generated by AI-PEOPLE, AI-CROWD andAI-OVERCROWD can be stored on board on the SD card of the camera through AI-Dashboard embedded.

The data can be visualized in two different ways:

n In tabular form, as a sequence of events. In this case (optionally) a sequence of

images associated to the events is available (not for AI-CROWD).

n The graphics related to the events generated by the plugins, with the possibility to

personalize the time interval and the time resolution.

11 / 153

AI-Dashboard embedded for data management

AI-Dash - configuration overview

Fig. 1: Sequence of events

Fig. 2: Graphic

AI-Dash - configuration overview

The dashboard in general is devided into the following sections:

n The main menu on top

n The live view area on the left

n The parameter section on the right

12 / 153

Fig. 3: Overview of the dashboard

Menu configuration

AI-Dashboard embedded for data management

AI-Dash - configuration overview

Fig. 4: Menu configuration

Note: Any changes made via AI-Config will only be applied to the application after the configuration has

been sent using the function in this panel.

The following functions are available:

Send configuration: the configuration will be send and stored to the application.

Reload configuration: the current configuration will be loaded from the application .

Save on file: The configuration can be downloaded an saved as file in JSON format.

Load from file: The saved configuration can be loaded from a file in JSON format.

Test: sends a test event to all the enabled channels in order to verify that the configuration of the channels

has been successful. Once clicked, simply click on the "Test" button in the window that will appear next. To exit the test mode, simply click anywhere on the screen.

13 / 153

AI-Dashboard embedded for data management

AI-Dash - configuration overview

Menu Administration

The administrator password to access the dashboard is the same as the cameras administrator password.

The default passwords for the app-administration are:

for AI-RETAIL: AIRetail3

for AI-BIO: AIBiolight

for AI-SECURITY: AISecurity3

for AI-TRAFFIC: AITraffic

Fig. 5: Menu Administration

The following functions are available:

Change configurator password: a configurator can load a configuration but change the parameters

Change admin password: an administrator can fully edit all parameters.

Menu License

Licensing is available in MxManagementCenter only .

14 / 153

Highlight objects in the foreground

Fig. 6: Highlight objects in foreground

AI-Dashboard embedded for data management

AI-Dash - configuration overview

1. Activate Highlights objects in the foreground to verify if the configuration of the low level parameters is

correct.

15 / 153

AI-Dashboard embedded for data management

AI-Dash - configuration overview

Scheduler

16 / 153

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Fig. 7: Scheduler

In many real installations, applications do not always need to be active. It may be required, for example, to enable the processing only from Monday to Friday, or every day at a certain time interval.

For this reason AI-RETAIL can be scheduled by configuring the periods in which they must be active and those in which they do not.

AI-Dash - Administrator parameters

For more experienced users, it is also possible to change the administrator parameters.

In this section, you can modify the low-level parameters that are required for background updating and extraction of the foreground mask. It is generally suggested that you do not change these parameters. However, the modification of these requires significant experience, so it was decided to protect this configuration with a password.

The administrator password to access the dashboard is the same as the cameras administrator password.

The default passwords for the app-administration are:

for AI-RETAIL: AIRetail3

for AI-BIO: AIBiolight

for AI-SECURITY: AISecurity3

for AI-TRAFFIC: AITraffic

Fig. 8: Administrator login with password

17 / 153

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Face detection (AI-BIO only)

Fig. 9: AI-BIO Face detection

Scaling factor: Growing factor of the window for the face detection (default 1,1). By increasing the value of

this parameter (max 2,0) you will make the algorithm faster but on the other hand it will became less sensible. Vice versa, by decreasing this value (min 1,01) the algorithm will become more sensible but also slower.

Number of classification stages: (default 25): Decreasing this value (it’s suggested to not set it less than 18), the algorithm sensitivity is increased, but also the false positive rate is increased.

Minimum number of rectangles:Minimum number of rectangles to consider an object as a detected face (default 1 - maximum sensibility). Decreasing this value, the algorithm sensitivity is increased, but also increase the false positive (min 1) rate is increased. On the other hand, if this value is excessively increased, the miss rate may increase (it’s suggested to not go further the value 10).

Shift step: Shift in pixels of the window for the face detection (default 2). Decreasing this value, the algorithm sensitivity and the processing time are increased(min 1). On the other hand, increasing this value, the sensitivity and the processing time may be reduced (it’s suggested to not go further the value 10).

18 / 153

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Gaussian filtering

Fig. 10: Gaussian filtering

Image pre-processing by gaussian filtering eliminates the acquisition noise on the image and makes subsequent operations for object detection easier and more effective. The

default kernel is 3x3, while other possible values are 5x5 and 7x7.

Gaussian filtering can also deactivated.

19 / 153

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Background

Fig. 11: Background

The background settings allow modeling and updating the background by setting the entry time of an object in the background

The output is an image in the YUV420 color space which represents the static part of the scene shot; it is then used to determine the dynamic part of the current frame, that is the foreground mask.

Update latency (s): Time period in seconds after a change in the scene must definitely become part of the background

Threshold (YUV): A comparison is made between the current frame and the background image of the previous instant: if the pixel of the frame is "close" to the corresponding pixel of the background, then it is not a foreground pixel; otherwise, that pixel will be white in the foreground mask. The comparison is made separately on each of the three YUV channels

Fig. 12: Example background extraction using a threshold for each of the three YUV channels.

20 / 153

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Update type: By specifying «Accurate (grayscale)» or «Accurate (color)» as background update type, it is

possible to use a state of the art self learning algorithm for extracting the foreground mask. The «grayscale» version uses only the Y color channel, while the «color» version uses all the channels; of course, the first is more efficient, while the second is more effective. Moreover, the shadow removal can be enabled only with the color version.

Morphological filtering

Fig. 13: Morphological filtering

Application of morphological erosion, dilation and another erosion operators to improve the foreground mask

Enable erosion (noise filtering): eliminates the spurious white pixels caused by noise

Enable dilation: fills holes and reinforces the union of weakly connected regions.

Enable erosion: allows to recover the original size of the objects.

It is possible to choose the shape of the kernel to be used (rectangular, diamond, octagon, disk), as well as the dimension in terms of width and height (rectangular) or radius (diamond, octagon, disk).

21 / 153

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Tracking (AI-BIO, AI-SECURITY only)

Fig. 14: Object tracking

Fig. 15: Tracking (AI-BIO and AI-SECURITY only)

Tracking of objects in different frames depending on the position in the image

The objective is to find the correspondence between the detected object to the preceding frame (t-1) and

the blob identified at the current frame (t), solving, in this way, problems related to occlusions (for example

trees)

Maximum radius: Maximum movement of an object between two successive frames. A too small value may cause frequent switches of the ID, while a too large value may cause the assignment of the same ID to different objects. The value is expressed as a fraction of the frame diagonal.

Max ghost time (ms): Maximum time (in milliseconds) for which a detected object can assume the status of ghost, namely it can be stored and retrieved in case of occlusion.

22 / 153

Small objects filtering (AI-SECURITY only)

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Fig. 16: Small objects filtering (AI-SECURITY only)

Elimination of blobs that are too small, too large or abnormally shaped based on pixel dimensions

Use aspect ratio: Check to activate the aspect ration settings. The settings allow to detect, for example, only

people or just cars.

Minimum Aspect Ratio: define the minimum value of the relationship between height and width.

Maximum Aspect Ratio: define the maximum value of the relationship between height and width.

Enable filtering: Check to activate the filtering settings.You can define minimum and maximum values for

the height and width of a blob by drawing a couple of rectangles on the image.

Maximum width and height: define the maximum value of the object size.

Miniumum width and height: define the minimum value of the object size.

23 / 153

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Filtering actual size (AI-SECURITY only)

To use this filter it is necessary first to calibrate the camera and the algorithm, to be able to

calculate the relation that allows to deduce the real dimensions of an object starting from the pixel

dimensions (see Camera Calibration (AI-SECURITY only), p. 25)

Fig. 17: Filtering actual size (AI-SECURITY only)

This filter allows the elimination of blobs that are too short or too tall based on actual size

Enable filtering: Check to activate the filtering settings.You can define minimum and maximum values for the height and width of a blob.

Maximum height: define the maximum height of a blob.

Minimum height: define the minimum height of a blob.

24 / 153

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Camera Calibration (AI-SECURITY only)

The camera calibration has to be done before filtering the actual size (see Filtering actual size (AI-

SECURITY only), p. 24)

Fig. 18: Camera Calibration (AI-SECURITY only)

This filter allows the elimination of blobs that are too short or too tall based on actual size.

Camera height (m): mounting height of the camera in meters.

Horizontal angle: cameras horizontal angle of view in degree. It is available on the datasheet of a fixed focal

cameras, to be calculated for varifocal cameras.

Vertical angle: cameras vertical angle of view in degree. It is available on the datasheet of a fixed focal cameras, to be calculated for varifocal cameras.

25 / 153

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Algorythm parameters (AI-SECURITY) only

Algorithm calibration

Algorythm calibration provides a collection of samples to train an algorithm that calculates the actual dimensions from those in pixels

Fig. 19: Algorythm calibration (AI-SECURITY only)

This filter allows the elimination of blobs that are too short or too tall based on actual size.

Show training data: Check to show trainiung data in the preview image.

Rotation (degrees): Camera rotation in reference to the horizontal plane.

Add element: ask a person of known height to move in different positions in the scene and at different

distances from the camera. Drawing a rectangle around the person every time he stops.

Delete element: Click to delete the selected element.

Height (m):Height of the element in meters.

26 / 153

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Shadow removal(AI-SECURITY only)

The algorithm for shadow removal is based on the analysis of the chromaticity difference between the background and the current frame, since the shadows typically make the pixels darker.

Fig. 20: Shadow removal (AI-SECURITY only)

Enable shadow removal: Check to activate the shadow removal settings.

Min fg-to-bg brightness ratio: Decreasing this value means the algorythm gets mor sensitive.

Max fg-to-bg brightness ratio: Increasing this value means the algorythm gets mor sensitive.

Max hue difference: Increasing this value means the algorythm gets mor sensitive and therefore removes

also strong shadows.

Max saturation increase: Increasing this value means the algorythm gets mor sensitive and therefore removes also strong shadows.

27 / 153

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Brightness control

Fig. 21: Brightness control

When sudden changes in brightness occur in the scene, the difference between the current frame and the background instantly becomes very high, generating a lot of noise on the foreground mask. The detection of this abnormal situation allows application to stop for a few moments the processing, allowing the background to automatically adapt to the brightness of the scene change.

For efficiency reasons the algorythm works on a grid built on the image and evaluates the differences in brightness only in grid intersections.

28 / 153

Performance

Fig. 22: Performance

AI-Dashboard embedded for data management

AI-Dash - Administrator parameters

Performance optimizations to make the algorithms more efficient.

Spatial decimation: consists in reducing the resolution at which the algorithm processes images. It is possible to reduce the size by a factor of 2 or 4, processing an image that is respectively a quarter or a sixteenth compared to the initial one.

Temporal decimation: allows to "discard" some frames, processing a picture every K milliseconds.

ROI: allows to perform the image processing only in the region drawn by the user.

Blob detection (AI-SECURITY only)

Fig. 23: Blob detection (AI-SECURITY) only

Tight bounding box (horizontal): consists in reducing the horizontal dimension of the bounding box by

centering it with respect to the centroid.

Tight bounding box (vertical): consists in reducing the vertical dimension of the bounding box by centering it with respect to the centroid.

Blob number limit: allows to limit the number of blobs detected by the plugin in a single frame.

29 / 153

AI-Dashboard embedded for data management

Event notification

Stream

Fig. 24: Stream

Ability to process a rotated image compared to that acquired by the camera. This operation, however, it may be interesting in the case where, for example, you want to install a camera in portrait mode, so as to take advantage of the horizontal opening angle of the camera to frame a person standing.

Device name:change the name of the stream

Rotation (degrees): image can be rotated by 90 °, 180 ° and 270 °.

Event notification

All AI-Apps can notify each event simultaneously to multiple recipients. You can enable and configure each recipient in the specific section of the events panel.

You can also specify for each event the channel on which you want to be notified. In the configuration section it is possible to enable the sending of only the desired events. This way you can completely customize the events sending. You can choose which event to send for each channel.

30 / 153

AI-Dashboard embedded for data management

Event notification

AI-RETAIL Events

Counting event is generated every time a person crosses a people counting sensor. The event gives

information about the number of persons which crossed the sensor simultaneously and related to the total number of crossings counted by the sensor since the last reset. It can be sent with and without images.

Aggregate event is generated when the number of persons (IN-OUT) is greater than a threshold configured by the user. Such event can be used as an alarm or like an advertisement of overcrowding, in case of a single entrance/exit gate. It can be sent with and without images.

Crowd event is generated periodically, with a period specified by the user during plugin configuration, giving an estimation of the average number of persons in the considered period. Such event can be used for collecting statistics about the retail shop. It can be sent ONLY without images.

Overcrowd event is generated when the estimated number of persons in the sensor is greater than a threshold configured by the user. Such event can be used as an alarm or like an advertisement of overcrowding. It can be sent with and without images.

Test event is generated by the user, clicking on the specific button on AI-Config. It can be used to verify the communication with the event collectors.

AI-BIO Events

Bio event is generated when a person, which face has been detected, leaves the scene. The event gives

information about the gender, the age category and the persistence time of each person in front of the camera. It can be sent with and without images.

Digital_Signage event is generated when persons are detected in front of the camera, after a minimum period of persistence. The event gives information about the average gender and age of the persons . It can be sent with and without images.

Test event is generated by the user, clicking on the specific button on AI-Config. It can be used to verify the communication with the event collectors.

AI-SECURITY Events

Sterile_Zone is generated when an intruder persists in a sterile zone. The event gives information about the

position of the object which generated the alarm. It can be sent with and without images.

Crossing_Line eventis generated when an object crosses a line. The event gives information about the position of the object which generated the alarm. It can be sent with and without images.

Intrusion_Pro eventis generated when an object crosses a multiple line. The event gives information about the position of the object which generated the alarm. It can be sent with and without images.

Lost event is generated when an object is abandoned or removed in a lost sensor. The event gives information about the position of the object which generated the alarm. It can be sent with and without images.

Loitering event is generated when a loitering behavior is detected in a loitering sensor. The event gives information about the position of the object which generated the alarm. It can be sent with and without images.

Test event is generated by the user, clicking on the specific button on AI-Config. It can be used to verify the communication with the event collectors.

31 / 153

AI-Dashboard embedded for data management

Event notification

AI-TRAFFIC Events

Sterile_Zone is generated when an intruder persists in a sterile zone. The event gives information about the

position of the object which generated the alarm. It can be sent with and without images.

Crossing_Line eventis generated when an object crosses a line. The event gives information about the position of the object which generated the alarm. It can be sent with and without images.

Intrusion_Pro eventis generated when an object crosses a multiple line. The event gives information about the position of the object which generated the alarm. It can be sent with and without images.

Test event event is generated by the user, clicking on the specific button on AI-Config. It can be used to verify the communication with the event collectors.

32 / 153

Image saving options

AI-Dashboard embedded for data management

Event notification

Fig. 25: Images saving options

Embed metadata:activate to enable the sending of annotated images (with sensors and bounding boxes for

example) associated to the events.

Line thickness: specify the thickness of bounding boxes and the font size of the superimposed strings.

Font size: specify the font size of the superimposed strings.

Modified bounding box:when enabled a bounding box is drawn, which allows to observe the object

orientation in the image.

Timestamp overlay: shows the date and hour overlay on the top right of the image.

Finally, since many event managers allows to send images in a time interval pre and post event, it is possible to specify the buffer size in frames and the time interval between consecutive frames saved in the buffer.

ATTENTION – The buffer size configuration and the temporal decimation with whom the frames are

stored impose a limit on the number to PRE and POST seconds of images that can be associated to

the events

33 / 153

AI-Dashboard embedded for data management

Event notification

Embedded AI-Dash

Fig. 26: Embedded AI-Dash

Enable AI-Dashboard embedded: activate to send events to AI-Dashboard embedded.

Embedded AI-Dash folder: folder in which theAI-Dashboard embedded database is created.

Maximum size:maximum size (in MB) that AI-Dashboard embedded can occupy on the device.

Send images:activate to send event images to AI-Dashboard embedded database

# Sec pre-event: Number of seconds of images before event.

# Sec post-event: Number of seconds of images after event.

34 / 153

External AI-Dash

AI-Dashboard embedded for data management

Event notification

Fig. 27: External AI-Dash

Enable send sending events: activate to send events to external AI-Dash.

IP: IP address of the server on which AI-Dash is installed (both server version and a cloud version).

Port: Port on which AI-Dashlistens.

AI-Dash ID: Once created on AI-Dash PRO the identifier related to your site and company, you can insert it in

this field. For more detail, please refer to the documentation of AI-Dash PRO.

Backward compatibility with AI-Dash: Enable this field if you have AI-Dash and not the new AI-Dash PRO (for more details please refer to the custom server notification in the following).

Send images:activate to send event images to AI-Dash database

# Sec pre-event: Number of seconds of images before event.

# Sec post-event: Number of seconds of images after event.

ATTENTION! To receive events, it may be necessary to disable the firewall

35 / 153

AI-Dashboard embedded for data management

Event notification

Wisenet WAVE

Fig. 28: Wisenet WAVE

Enable send sending events: activate to send events to Wisenet Wave.

IP: IP address of Wisenet WAVE VMS.

Port: Port number of the Wisenet WAVE VMS.

Username: Username to authenticate the Wisenet WAVE VMS.

Password:Password to authenticate to the Wisenet WAVE VMS.

Use HTTPS:activate to send events through https

The event sending to Wisenet WAVE is not supported for Crowd events

36 / 153

Hanwha SSM

AI-Dashboard embedded for data management

Event notification

Fig. 29: Hanwha SSM

Enable send sending events: activate to send events to Hanwha SSM.

IP: IP address of the server on which SSM is installed

Port: Port number of the SSM.

Device GUID: device identifier to read on SSM.

Set the server timezone:SSM server timezone.

The sending of events to Wisenet SSM is not supported for Crowd events.

37 / 153

AI-Dashboard embedded for data management

Event notification

Text Sender Configuration

This mechanism makes the app integrated with the Wisenet NVR.

Fig. 30: Text sender configuration

Enable send sending events: activate to send events.

IP: IP address of the server on which AI-Dash is installed in both the server version and a cloud version.

Port: Port number.

Path: Path for the POST to the receiving server;.

MIME type: MIME Type with which the message will be transmitted.

charset: Character set for the message text.

Use URL Encode: indicates that the message is encoded using URL Encode for sending.

Message Format: message texrt sent to the server. Thes palceholders can be used in the message text

n event name: %e

n device name: %d

n sensor name: %s

n date: %t (format DD / MM / YYYY

Use URL Encode: indicates that the message is encoded using URL Encode for sending.

38 / 153

AI-Dashboard embedded for data management

The sending of text events is not supported for Crowd events.

Digital output

Event notification

Fig. 31: Digital output

Enable send sending events: activate to send event via a digital output.

Single pulse duration (ms): duration of a single pulse in miliseconds.

Pulse Interval (ms): Time in ms between two pulses.

Number of pulses: Number of pulses sent through the alarm out port.

Device: Device on which the application is running.

Pin: Pin you want to use on the device.

Sending of events to digital inputs is not supported for Crowd events.

39 / 153

AI-Dashboard embedded for data management

Event notification

HTTP I/O

Fig. 32: HTTP I/O

Enable send sending events: activate to send event via generic I / O (for example to call the CGIs of the

Wisenet NVR).

IP: IP address of the remote I / O.

Port: port on which is listening on the remote I / O.

Path: Path of the remote I / O.

Username: user name to connect to the remote I / O.

Password: Password to connect to the remote I / O.

Parameters: query string with all the required parameters. The format allows to add information about the

event. It’s necessary to add the following tags to the message

n event name: %e

n device name: %d

n sensor name: %s

n date: %t (format DD / MM / YYYY

Use HTTPS: : if checked, send through HTTPS.

40 / 153

AI-Dashboard embedded for data management

Event notification

example to set 10 seconds of duration of an alarm on the Hanwha NVR by using Hanwha SUNAPI:

http://172.16.35.160/stwcgi/io.cgi?msubmenu=alarmoutput&action=set&AlarmOutput.1.IdleState=Normally-

Open&AlarmOutput.1.ManualDuration=10s



41 / 153

AI-Dashboard embedded for data management

Event notification

Sending event to E-mail

Fig. 33: Sending event to E-mail

Enable send sending events: activate to send event via email.

Sender: e-mail address of the sender.

Username:sender's user name for SMTP server access.

Password: sender's password for SMTP server access.

SMTP Server: address of the SMTP server.

SMTP port: port number of the SMTP server.

Recipients: You can enter multiple email addresses separated by a semicolon.

# Sec pre-event: Number of seconds of images before event.

# Sec post-event: Number of seconds of images after event.

The sending of events by e-mail is not supported for Crowd events

42 / 153

Sending event to Milestone

AI-Dashboard embedded for data management

Event notification

Fig. 34: Sending event to Milestone

Enable send sending events: activate to send event to Milestone XProtect®

IP server: IP address of the server on which you installed Milestone XProtect®, both server version and a

cloud version.

Server port:Port number to listen for Milestone XProtect® events.

IP device: IP address of the device.

Timezone: Timezone of Milestone XProtect® servers.

The sending of events to Milestone XProtect® is not supported for Crowd events.

43 / 153

AI-Dashboard embedded for data management

Event notification

Sending event to Arteco EVERYWHERE

Fig. 35: Sending event to Arteco EVERYWHERE

Enable send sending events: activate to send event to Arteco EVERYWHERE.

IP: IP address of the server on which you installed Milestone Arteco EVERYWHERE, both server version and a

cloud version.

Server port: Port number to listen for Arteco EVERYWHERE.

Username: Username for login to Arteco EVERYWHERE server.

Password: Password for login to Arteco EVERYWHERE server.

Number of output: Output number associated with the event.

The sending of events to Arteco EVERYWHERE is not supported for Crowd events.

44 / 153

Sending event to Arteco NEXT

AI-Dashboard embedded for data management

Event notification

Fig. 36: Sending event to Arteco NEXT

Enable send sending events: activate to send event to Arteco NEXT.

IP: IP address of the server on which you installed Milestone Arteco NEXT, both server version and a cloud

version.

Server port: Port number to listen for Arteco NEXT server.

Username: Username for login to Arteco NEXT server.

Password: Password for login to Arteco NEXTserver.

Connector ID: Identification of the connector defined in Arteco NEXT for sending event notifications.

Camera ID: Identification of the connector defined in Arteco NEXT for sending event notifications.

Description: Information that will be displayed in Arteco NEXT related to the application of video analysis.

The sending of events to Arteco NEXT is not supported for Crowd events.

45 / 153

AI-Dashboard embedded for data management

Event notification

Sending event to Avigilon POS

Fig. 37: Sending event to Avigilon POS

Enable send sending events: activate to send event to Avigilon POS.

Port: Port number on which the Avigilon server is listening.

Beginning event string: characters at the beginning of the event.

Ending event string: characters at the end of the event.

The sending of events to Avigilon POS is not supported for Crowd events.

46 / 153

Sending event to FTP server

AI-Dashboard embedded for data management

Event notification

Fig. 38: Sending event to FTP server

Enable send sending events: activate to send event to a FTP server.

IP: IP address of the FTP server.

Port: port number of the FTP server.

Username: Username to authenticate to the FTP server.

Password: Password to authenticate to the FTP server.

Path of destination: Path, defined from the root folder, FTP, to transfer the files to the server.

Send images: check to include images in the event sent.

47 / 153

AI-Dashboard embedded for data management

Event notification

Remote server

Fig. 39: Sending event to Remote server

Enable send sending events: activate to send event to remote server.

IP Server: IP address of the remote server.

Port: port number of the remote server.

Path: Path for the POST to the receiving server.

Send json as “form-data”: Enables url encoding for the message sent.

Send images: check to include images in the event sent.

# Sec pre-event: Number of seconds of images before event.

# Sec post-event: Number of seconds of images after event.

Backward compatibility withAI-Dash: Enable this field if you want to receive events compliant with AI-Dash

and not the new AI-Dash PRO (for more details please refer to the custom server notification in the following).

48 / 153

AI-Dashboard embedded for data management

Event notification

Input by web request

The event notification triggering through web request event_switch.cgi is available for all applications on all platforms.

Fig. 40: Input by web request

Use event activation/deactivation via web request: activate to manage the input via web request.

Password: Required to avoid fraudulent activation/deactivation.

Events initially enabled: If enabled, the events are initially activated and in presence of web inputs are

inhibited. Otherwise, events are initially inhibited and will be activated in presence of web inputs.

Behaviour: Possible values are: timed or on/off. An on/off input enables/disables the sending of events on the rising edge. A time input enables / disables the sending of events for a certain time interval, specified by the "Switch duration" parameter.

EXAMPLE:

disable events (because they are initially enabled) on a device with ip 192.168.1.1 and password «foo». If the

behaviour is Timed, the events will be disabled for Switch duration ms

http://192.168.1.1:8080/local/AIRetail3/event_switch.cgi?mode=toggle&password=foo

49 / 153

AI-Dashboard embedded for data management

Event notification

HTTP request sequence

The plugin also allows to send sequences of HTTP requests, interspersed with a configurable time interval. As an example, you may think to move a PTZ camera on different presets or create your custom sequence to drive remote I/O devices. It is possible to configure an unlimited number of requests in the sequence.

Fig. 41: Input by web request

Enable sending events: activate end events via HTTPrequest sequence.

Suspend elaboration during sequence: Enable it to suspend the elaboration during the sequence.

Http(s) URI: The path of the HTTP(s) request.

Time before next item (s): Time interval in seconds to call the next request in the sequence.



50 / 153

AI-Dashboard embedded for data management

Custom server compliant event notification

Event notification

51 / 153

AI-Dashboard embedded for data management

Event notification

Custom server compliant with AI-Dash

Events without images: POST with application/json

POST /pathOnServer HTTP/1.1

Accept: application/json

Host: 172.16.35.75:8080

Content-Type: application/json

Content-Length: 157

json_data:{"id_source": " people_sensor", "sensor_id": 4, "sensor_name":

"S1", "event_type": "Counting", "mac_address": "b827ebc42838", "timestamp":

1510761996, "actual_count": 35, "people_number": 1, "dash_id":

“Site#Company"}

52 / 153

AI-Dashboard embedded for data management

Event notification

Events with images: POST with custom multipart/x-mixed-replace

POST /pathOnServer HTTP/1.1

Accept: */*

Host: 172.16.35.75:8080

Content-length: 3844500

Content-Type: multipart/x-mixed-replace; boundary=gc0p4Jq0M2Yt08jU534c0p

--gc0p4Jq0M2Yt08jU534c0p

Content-Type: application/json

Content-Length: 157

{"id_source": " people_sensor", "sensor_id": 4, "sensor_name": "S1",

"event_type": "Counting", "mac_address": "b827ebc42838", "timestamp":

1510761996, "actual_count": 35, "people_number": 1, "dash_id":

“Site#Company"}

--gc0p4Jq0M2Yt08jU534c0p

Content-Type: image/jpeg

Content-length: 11146

JPEG DATA

…

53 / 153

AI-Dashboard embedded for data management

Event notification

Custom server compliant with AI-Dash PRO

Events without images: POST with application/json

POST /pathOnServer HTTP/1.1

Accept: application/json

Host: 172.16.35.75:8080

Content-Type: application/json

Content-Length: 157

{"id_source": " people_sensor", "sensor_id": 4, "sensor_name": "S1",

"event_type": "Counting", "mac_address": "b827ebc42838", "timestamp":

1510761996, "actual_count": 35, "people_number": 1, "dash_id":

“Site#Company"}

The headers are case insensitive as stated in HTTP standard. The number of images is variable. This

value will be different based on images and event size. Note that the transfer encoding is set to

chunked: HTTP 1.1 support is required. On the raw socket each chunk will follow a row with the

number of bytes and followed by a new row.

See https://tools.ietf.org/html/rfc7230#section-4.1 for details on chunked encoding.

See https://www.w3.org/Protocols/rfc1341/7_2_Multipart.html for details on multipart/mixed content type.

54 / 153

AI-Dashboard embedded for data management

Event notification

Events with images: POST with custom multipart/x-mixed-replace

POST /www/prova.php

HTTP/1.1Host: 172.16.35.28:80

Accept:

text/html,application/json,application/xhtml+xml,application/xml;q=0.9,*/*;-

q=0.8

User-agent: axia_http_client/2.6.1

Content-Type: multipart/mixed;boundary=gc0p4Jq0M2Yt08jU534c0p

Transfer-encoding: chunked

--gc0p4Jq0M2Yt08jU534c0p

Content-Disposition: form-data;

name="json"

Content-Type: application/json

Content-Length: 157

{"id_source": " people_sensor", "sensor_id": 4, "sensor_name": "S1",

"event_type": "Counting", "mac_address": "b827ebc42838", "timestamp":

1510761996, "actual_count": 35, "people_number": 1, "dash_id":

“Site#Company"}

--gc0p4Jq0M2Yt08jU534c0p

Content-Disposition: form-data;

name="jpeg"

Content-Type: image/jpeg

Content-length: 60155

################# BINARY JPEG DATA (60155 bytes total) #################

--gc0p4Jq0M2Yt08jU534c0p--

55 / 153

AI-Dashboard embedded for data management

AI-Dash - troubleshooting

Custom server –JSON event format

JSON

field

id_source string Name of the device, specified in the plugin configuration All

event_

type

timestamp string Value which represents the number of seconds passed since 00:00 of the 1st

sensor_id integer Id associated to the sensor which generated the event All

sensor_

name

mac_

address

dash_id string An identifier of the site and the company, specified in the plugin

people_

Value

type

string Type of event. It can assume values: Counting, Aggregate, Crowd, Overcrowd All

string Name associated to the sensor which generated the event All

string MAC address of the device that generated the event All

integer For Counting events, represents the number of persons crossing

Description Type of

events

All

January 1970 UTC (for instance, a Unix timestamp)

All

configuration

All

number

actual_

count

period integer For Crowd events, interval between two consecutive events Crowd



integer For Counting events, represents the total number of persons counted by the

simultaneously the sensor. For Aggregate events, represents the current IN-

OUT value. For Crowd and Overcrowd events, represents the number of

estimated persons in the sensor.

sensor since the last reset. For Aggregate events, represents the current IN-

OUT value.

Counting,

Aggregate

AI-Dash - troubleshooting

In case of low bandwidth (e. g. because of huge network-load or undersized systems) or the camera is overloaded, the live screen may be loading slowly or not to show live. In addition, some browsers may activate filters that block streaming by default (usually Chrome, Firefox and Safari do not have locks).

In these cases:

n Reloading the page and wait for the live image

n Use a different web-browser

If image is displayed is green try to perform the following operations:

56 / 153

AI-Dashboard embedded for data management

AI-Dash - troubleshooting

n Restart the camer, or alternatively reset to the initial settings (except those related to the application);

n Verify that the latest firmwareis installed on the camera

n Contact technical support (see Support, p. 5)

57 / 153

AI-SECURITY

AI-SECURITY is a bundle including three different products, simultaneously installed on board of your camera.

n AI-INTRUSION: Intrusion detection in sterile zone and virtual line crossing

n AI-LOST: Abandoned or removed objects detection

n AI-LOITERING: Loitering detection in forbidden areas

AI-SECURITY - camera positions

n Make sure the size of the target (person, object, animal, vehicle) is at least 10x10

pixels.

n If necessary, the camera should be mounted with external illuminators, to

distinguish the targets with natural or artificial illumination.

n The camera should be mounted at a height between 3 and 5 meters

The precision of the plugins can be reduced if there are occlusions, waving objects,

n

vehicles which project light in interest areas and any other noise that continuously modifies the image

59 / 153

AI-SECURITY

AI-INTRUSION

Fig. 42: Camera positions

Intrusion sensors, p. 63

AI-INTRUSION

AI-INTRUSION is a video analytic app that is able to detect intruders in indoor and outdoor environments; thus, the environmental conditions will affect the performance of the application.

The accuracy to be expected is under ideal environmental and installation conditions

n Recall: 95%

AI-INTRUSION

60 / 153

AI-SECURITY

AI-INTRUSION

Environment conditions

The position of the camera and the environmental conditions affect the performance of the application. Performance is best under the following conditions:

n The image must not present flickering, severe noise or artifacts.

n Image must have a resolution of 640x360, 640x480, 320x180, 320x240.

n Rotating (PTZ) security cameras are supported only if they are not moved when the application is

enabled. If the camera is moved, the application must be reconfigured.

n Absence of occlusions (E. g. Trees, pillars, buildings, furniture elements etc.) that do not allow to see

the people.

n Absence of conditions of high crowding or stopped people that do not allow to count the individuals.

n Absence of stationary or slow-moving people for long periods in the counting area (e.g. Sales people

that encourage customers to enter).

There must be no fog, clouds or other moving objects whose appearance is similar to the target in the

n

areas of interest.

n Camera lens must not be dirty, wet or covered in rain or water drops. Camera lens must not be

steamy.

n Absence of "waving objects" (e.g. Meadow with tall grass, trees, sliding doors, etc.) or any other type

of disturbance that causes the continuous modification of the images (moving pixels) in the areas of

interest.

n Camera placement must be stable and solid in a way that wind or external disturbances of other types

will cause movement of the camera that appears on the image.

n Absence of vehicles with lights projected in areas of interest.

n Correct exposition of the camera: camera must not be in backlight, the framed area must not have

heterogeneous illumination, i.e. partially indoor or partially outdoor. In general, no areas to be

monitored must be almost white or almost black, i.e. the dynamic range must be sufficient to

correctly show detail of objects in the image. If necessary, the camera must be installed with external

illuminators that make it possible to distinguish the people in all natural or artificial lighting

conditions.

n The people must have a sufficient dissimilarity from the background, i.e. there is no explicit

camouflage, where the people are similar to the background in color and texture. Sufficient

dissimilarity means at least a color difference of at least 5% or a brightness difference of at least 10%.

The target must stay in the interested area for a time of at least 1 second.

n

The target must have a minimum area of 100 pixels.

The target must move at a maximum speed of half their width on the image per frame. For example, a

n

target that is 40 pixels wide at 10 frames per second must move at a speed of no more than 20 pixels per frame, that is 200 pixel per second.

The scene must be predominantly non-reflective.

n

n No hard lights must be present that cast shadows in a way that the background brightness is reduced

to less than 50% of the original value in the image.

61 / 153

AI-SECURITY

AI-INTRUSION

In case of thermal cameras, the image must be not coloured but in grayscale (white for “hot” pixels,

n

black for “cold” pixels). The camera, thermal or monocular, must be always configured in order to avoid continuous changes of brightness.

AI-INTRUSION - target size

62 / 153

Intrusion sensors

AI-SECURITY

AI-INTRUSION

63 / 153

AI-SECURITY

AI-INTRUSION

64 / 153

AI-SECURITY

AI-INTRUSION

Fig. 43: Configuration of AI-INTRUSION Intrusion sensors

The configuration section provides the following functions:

Add Sensor: Click this button to draw the area of interest directly on the live image on the left. The area of interest it’s a polygon with no limits to the number of sides.

Remove sensor: Click this button to remove the selected sensor from the configuration.

Redraw sensor: Click this button to redraw the current sensor. The current area of interrest will be deleted.

ID sensor: define a numeric IDfor the sensor.

Sensor name: this name uniquely identifies the main counting sensor (green arrow); is used to generate

counting events, sent, for example, to AI-Dash.

Confidence: A small value will make the algorithm very sensitive, instead with a value too large the algorithm could not generate the alarms.

Inhibition (s): Inhibition time in seconds of the sensor after an alarm has been generated. If an alarm is generated by the same sensor before the inhibition time is passed, it will be ignored by the system.

Latency alarm (s): Minimum intrusion time (seconds of permanence in the area of interest) before an alarm is generate. Time in seconds. Subjects who stay in the area of interest for less time than the set latency, won’t generate any alarm.

Sensor type: there are two types of sensors:

Impulsive: generates a single event for the whole duration of the intrusion.

n

Levels: generates several types of event: beginning of the intrusion, intrusion continuation (every

n

“Inhibition” seconds) and end of intrusion.

End time intrusion: after this amount of seconds, if none is in the level sensor, an event of “end of intrusion” will be sent.

65 / 153

AI-SECURITY

AI-INTRUSION

Crossing the line

66 / 153

AI-SECURITY

AI-INTRUSION

67 / 153

AI-SECURITY

AI-INTRUSION

Fig. 44: Configuration of AI-INTRUSION Crossing line sensors

The configuration section provides the following functions:

Add Sensor: Click this button to draw the area of interest directly on the live image on the left. The area of interest it’s a polygon with no limits to the number of sides.

Remove sensor: Click this button to remove the selected sensor from the configuration.

Redraw sensor: Click this button to redraw the current sensor. The current area of interrest will be deleted.

ID sensor: define a numeric IDfor the sensor.

Sensor name: this name uniquely identifies the sensor, it is used to generate events to be sent for example

to AI-Dash.

Crossing line pre confidence: confidence relative to the object before it crosses the line ( pre alarm).

Crossing line post confidence: confidence relative to the activation of the alarm ( crossing the line) on a

object already considered in a pre-alarm state.

Crossing line pre latency: time of latency of an object that is in the scene before it crosses the line (prealarm). Time in seconds.

Crossing line post latency: time of latency of an object already considered in pre-alarm state that it spends in the scene after it crosses the line. Time in seconds.

68 / 153

AI-SECURITY

AI-INTRUSION

Multiple crossing lines

A multiple crossing line sensor is an aggregate sensor inside the scene consisting of a set of crossing lines (see Crossing the line, p. 66). If the subject crosses all the lines specified in the scene, the alarm will be generated.

69 / 153

AI-SECURITY

AI-INTRUSION

70 / 153

AI-SECURITY

AI-LOITERING

Fig. 45: Configuration of AI-INTRUSION - Multiple crossing line sensors

The configuration section provides the following functions:

Add aggregate sensor: Click this button to draw the area of interest directly on the live image on the left. The area of interest it’s a polygon with no limits to the number of sides. The aggregate sensor can contain multiple crossing lines.

Remove aggregate sensor: Click this button to remove the selected aggregate sensor from the configuration.

ID sensor: define a numeric IDfor the aggregate sensor.

Sensor name: this name uniquely identifies the aggregate sensor, it is used to generate events to be sent for

example to AI-Dash.

Crossing time (s): maximum crossing time in seconds between two successive crossing lines.

It is required to add crossing line sensors within the aggregate sensor (See Crossing the line, p. 66).

AI-LOITERING

AI-LOITERING is a video analytic app that is able to detect loitering in indoor and outdoor environments; thus, the environmental conditions will affect the performance of the application, FTP servers and third party servers.

The accuracy to be expected is under ideal environmental and installation conditions

n Recall: 95%

Fig. 46: AI-LOITERING: configuration

71 / 153

AI-SECURITY

AI-LOITERING

Environment conditions

The position of the camera and the environmental conditions affect the performance of the application. Performance is best under the following conditions:

n The image must not present flickering, severe noise or artifacts.

n Image must have a resolution of 640x360, 640x480, 320x180, 320x240.

n Rotating (PTZ) security cameras are supported only if they are not moved when the application is

enabled. If the camera is moved, the application must be reconfigured.

n Absence of occlusions (E. g. Trees, pillars, buildings, furniture elements etc.) that do not allow to see

the people.

n Absence of conditions of high crowding or stopped people that do not allow to count the individuals.

n There must be no fog, clouds or other moving objects whose appearance is similar to the target in the

areas of interest.

n Camera lens must not be dirty, wet or covered in rain or water drops. Camera lens must not be

steamy.

n Absence of "waving objects" (e.g. Meadow with tall grass, trees, sliding doors, etc.) or any other type

of disturbance that causes the continuous modification of the images (moving pixels) in the areas of

interest.

n Camera placement must be stable and solid in a way that wind or external disturbances of other types

will cause movement of the camera that appears on the image

n Absence of vehicles with lights projected in areas of interest.

n Correct exposition of the camera: camera must not be in backlight, the framed area must not have

heterogeneous illumination, i.e. partially indoor or partially outdoor. In general, no areas to be

monitored must be almost white or almost black, i.e. the dynamic range must be sufficient to

correctly show detail of objects in the image. If necessary, the camera must be installed with external

illuminators that make it possible to distinguish the people in all natural or artificial lighting

conditions.

The target must have a sufficient dissimilarity from the background, i.e. there is no explicit

n

camouflage, where the target is similar to the background in color and texture. Sufficient dissimilarity means at least a color difference of at least 5% or a brightness difference of at least 10%.

n The target must stay in the interested area for a time of at least 5 seconds.

n The target must have a minimum area of 100 pixels.

n The target must move at a maximum speed of half their width on the image per frame. For example, a

target that is 40 pixels wide at 10 frames per second must move at a speed of no more than 20 pixels

per frame, that is 200 pixel per second.

n The scene must be predominantly non-reflective.

n No hard lights must be present that cast shadows in a way that the background brightness is reduced

to less than 50% of the original value in the image.

72 / 153

AI-SECURITY

AI-LOITERING

n In case of thermal cameras, the image must be not coloured but in grayscale (white for “hot” pixels,

black for “cold” pixels). The camera, thermal or monocular, must be always configured in order to

avoid continuous changes of brightness.

Installation constraints

A camera that can be used to detect loitering with AI-LOITERING must comply with the following installation restrictions (in addition to the respect of the environmental conditions):

It must be installed in such a way that the framed targets (people, vehicles, animals) have a minimum

n

area of 100 pixels.

If necessary, it must be installed with external illuminators that make it possible to distinguish the

n

targets in all natural or artificial lighting conditions.

73 / 153

AI-SECURITY

AI-LOITERING

Configuration of AI-LOITERING sensors

Fig. 47: Configuration of AI-LOITERING sensors

The configuration section provides the following functions:

Add Sensor: Click this button to draw the area of interest directly on the live image on the left. The area of interest it’s a polygon with no limits to the number of sides.

Remove sensor: Click this button to remove the selected sensor from the configuration.

Redraw the sensor: Click to delete the current sensor and draw a new one.

ID sensor: define an ID number for the sensor.

Sensor name: this name uniquely identifies the sensor.

Confidence: A small value will make the algorithm very sensitive, instead with a value too large the

algorithm could not generate the alarms.

74 / 153

AI-SECURITY

AI-LOST

Latency alarm (s): Minimum intrusion time (seconds of permanence in the area of interest) before an alarm

is generate. Time in seconds. Subjects who stay in the area of interest for less time than the set latency, won’t generate any alarm.

AI-LOST

AI-LOST is a video analytic app that is able to detect abandoned or removed objects in indoor and outdoor environments; thus, the environmental conditions will affect the performance of the application.

The accuracy to be expected is under ideal environmental and installation conditions

n Recall: 90%

Fig. 48: AI-LOST: configuration

75 / 153

AI-SECURITY

AI-LOST

Environment conditions

The position of the camera and the environmental conditions affect the performance of the application. Performance is best under the following conditions:

n The image must not present flickering, severe noise or artifacts.

n Image must have a resolution of 640x360, 640x480, 320x180, 320x240.

n Rotating (PTZ) security cameras are supported only if they are not moved when the application is

enabled. If the camera is moved, the application must be reconfigured.

n Absence of occlusions (E. g. Trees, pillars, buildings, furniture elements etc.) that do not allow to see

the people.

n Absence of conditions of high crowding or stopped people that do not allow to count the individuals.

n There must be no fog, clouds or other moving objects whose appearance is similar to the target in the

areas of interest.

n Camera lens must not be dirty, wet or covered in rain or water drops. Camera lens must not be

steamy.

n Absence of "waving objects" (e.g. Meadow with tall grass, trees, sliding doors, etc.) or any other type

of disturbance that causes the continuous modification of the images (moving pixels) in the areas of

interest.

n Camera placement must be stable and solid in a way that wind or external disturbances of other types

will cause movement of the camera that appears on the image

n Absence of vehicles with lights projected in areas of interest.

n Correct exposition of the camera: camera must not be in backlight, the framed area must not have

heterogeneous illumination, i.e. partially indoor or partially outdoor. In general, no areas to be

monitored must be almost white or almost black, i.e. the dynamic range must be sufficient to

correctly show detail of objects in the image. If necessary, the camera must be installed with external

illuminators that make it possible to distinguish the people in all natural or artificial lighting

conditions.

The target must have a sufficient dissimilarity from the background, i.e. there is no explicit

n

camouflage, where the target is similar to the background in color and texture. Sufficient dissimilarity means at least a color difference of at least 5% or a brightness difference of at least 10%.

n The target must stay in the interested area for a time of at least 5 seconds.

n The target must have a minimum area of 100 pixels.

n The target must move at a maximum speed of half their width on the image per frame. For example, a

target that is 40 pixels wide at 10 frames per second must move at a speed of no more than 20 pixels

per frame, that is 200 pixel per second.

n The scene must be predominantly non-reflective.

n No hard lights must be present that cast shadows in a way that the background brightness is reduced

to less than 50% of the original value in the image.

76 / 153

AI-SECURITY

AI-LOST

n In case of thermal cameras, the image must be not coloured but in grayscale (white for “hot” pixels,

black for “cold” pixels). The camera, thermal or monocular, must be always configured in order to

avoid continuous changes of brightness.

AI-LOST - target size

Installation constraints

A camera that can be used to detect loitering with AI-LOITERING must comply with the following installation restrictions (in addition to the respect of the environmental conditions):

It must be installed in such a way that the framed targets (people, vehicles, animals) have a minimum

n

area of 100 pixels.

If necessary, it must be installed with external illuminators that make it possible to distinguish the

n

targets in all natural or artificial lighting conditions.

77 / 153

AI-SECURITY

AI-LOST

Configuration of AI-LOST sensors

Fig. 49: Configuration of AI-LOST sensors

The configuration section provides the following functions:

Add Sensor: Click this button to draw the area of interest directly on the live image on the left. The area of interest it’s a polygon with no limits to the number of sides.

Remove sensor: Click this button to remove the selected sensor from the configuration.

Redraw the sensor: Click to delete the current sensor and draw a new one.

ID sensor: define an ID number for the sensor.

Sensor name: this name uniquely identifies the sensor.

Confidence: A small value will make the algorithm very sensitive, instead with a value too large the

algorithm could not generate the alarms.

78 / 153

AI-SECURITY

AI-LOST

Latency alarm (s): Minimum intrusion time (seconds of permanence in the area of interest) before an alarm

is generate. Time in seconds. Subjects who stay in the area of interest for less time than the set latency, won’t generate any alarm.

Configuration of AI-LOST Entrance areas



In order to reduce the number of false positives and to consider only the objects which enters from specific parts of the image, it is possible to draw an unlimited number of entrance areas.

Fig. 50: Configuration of AI-LOST entrance areas

The configuration section provides the following functions:

Add entrance area: Click this button to draw an entrance area of directly on the live image on the left. The entrance area is a polygon with no limits to the number of sides.

Delete entrance area: Click this button to remove the selected entrance area from the configuration.

79 / 153

AI-RETAIL3

AI-RETAIL3 is a bundle including three different products, simultaneously installed on board of your camera.

n AI-PEOPLE: People counting through gates

n AI-CROWD: Crowd estimation

n AI-OVERCROWD: Overcrowding detection for queue management



AI-RETAIL - camera positions

n The camera should be mounted with a reduced focal length and an horizontal field

of view in the range between 60° and 120°, chosen with respect to the gate.

n The camera must be mounted in a overhead position considering an 90° angle

measured to ground.

n The camera should be mounted at a height between 3 and 5 meters

n The precision of the plugins is maximum when people are recorded from the top

without distorsion on the sides

81 / 153

AI-RETAIL3

AI-PEOPLE

Fig. 51: camera position

Recomended distances

Camera height (m) Maximum gate width (m)

3 6

3,5 7,5

4 9

4,5 10

5 12



AI-PEOPLE

AI-PEOPLE is a video analytic plugin optimized to count people crossing a gate in typical retail scenarios. It generates events that can be managed by all the notification channels.

The accuracy to be expected is under ideal environmental and installation conditions

Indoor:

n Recall: 85%

n Precision: 95%

Outdoor:

n Recall: 85%

n Precision: 85%

82 / 153

AI-RETAIL3

AI-PEOPLE

Environment conditions

The position of the camera and the environmental conditions affect the performance of the application. Performance is best under the following conditions:

n The image must not present flickering, severe noise or artifacts.

n Image must have a resolution of 640x360, 640x480, 320x180, 320x240.

n Rotating (PTZ) security cameras are supported only if they are not moved when the application is

enabled. If the camera is moved, the application must be reconfigured.

n Absence of occlusions (E. g. Trees, pillars, buildings, furniture elements etc.) that do not allow to see

the people.

n Absence of conditions of high crowding or stopped people that do not allow to count the individuals.

n Absence of stationary or slow-moving people for long periods in the counting area (e.g. Sales people

that encourage customers to enter).

n There must be no other moving objects whose appearance is similar to the people in the areas of