CrowdCount

Use the CrowdCount module to control overcrowded areas (e.g., at transportation hubs), manage queues, respond to social distancing violations, etc.

Meet CrowdCount

How it works

At the core of the module, there is a convolutional neural network (CNN) trained to detect human figures in open spaces, such as railway platforms, airport terminals, etc. It processes every n-th frame in the incoming video stream to build annotations (bounding boxes) around detected figures; in between, the figures are tracked.

Human figure bounding boxes are then checked against the event rules, specifically:

• location within the Region of interest (ROI).
• the number of people in the ROI.
• time spent in it.

Depending on the result, a "Crowd" event is fired or terminated.

Camera and video requirements

The module performs best under the following conditions:

• Camera type: fixed or PTZ; en-face view shall be ensured as much as possible.
• Resolution: from 1280x720 (HD Ready) to QHD 2592x1944.
• Tilt: +15° to +75°
• Pan: -75° to +75°
• Roll -5° to +5°
• Height: from 2 to 5 meters above the ground level.
• Image quality within the region of interest: over 100 px/m
• Minimum human outline size: 2025 (45X45) at a crowd less than 15, 400 (20x20) at a crowd from 15 to 35.

Computational capacity

The computational complexity of the module is estimated as "Average"; the exact numbers may vary with video resolution and CPU parameters.

Export module settings

Export the module settings to a *.xml file, e.g., to pass it to qualified IREX personnel to find a solution to a problem with the module performance, should it happen. To do so:

1. Find the camera running the module in the User menu — Settings — Cameras, and click Video  analytics settings as shown:    

2. On the VIDEO ANALYTICS tab on the camera settings pane on the left, click the three-dot menu 



Export analytics configurations of the module — to export the module settings on the MAIN sub-tab.

Export analytics configurations of the model — to export the entire module configuration, including the rules, if any.

Enable CrowdCount

1. Find the camera to enable the module in the User menu — Settings — Cameras section and click Video analytics settings as shown.    

2. On the <module_name> list, click the name of the module.
3. Slide right the On/Off slider 
4. Click Apply at the bottom of the pane.

How to detect a crowd

Crowding may need a real-time response from the security or premises management teams.

1. Make sure the cameras monitoring potentially crowded places meet the system requirements.
2. Enable CrowdCount on those cameras and set it up to detect crowds

Remember, CrowdCount can run in parallel with any other module.

3. Build an alarm monitor having Analytics modules - CrowdCount - Crowds checkmarked.
4. Review the crowd alarms arriving onto the monitor.

Set up crowd detection

Under CALIBRATION  click Region of interest and build the part of the frame to analyze

To move a vertex, point the cursor onto it and drag and drop it.
To delete a vertex, double-click it.
To add a vertex onto an edge, double-click where the vertex should be.
To move the region as a whole, place the cursor inside it and drag-and-drop.

Under ADDITIONAL SETTINGS, define:

• Count people by: select Silhouette if the expected number is moderate enough, or Top of head for big crowds. With Top of head selected, Save people's position gets enabled; switch it on to make it possible to calculate "on the fly" the number of people in a user-defined area in the media player while a video is playing.
• People in crowd - the number of people to recognize a crowd; an integer between 3 and 35.
• Crowd after ... s - time limit for too many (above the Number of people in a crowd) people to stay within the Region of interest to detect a crowd; integer between 15 and 600, in seconds. Must be 3 times longer than the Recount every ... s
• to make the detection of a crowd confident enough.
• Recount every ... s - check for crowd period, from 5 to 60 seconds. Must be at least 3 times shorter than the Crowd after ... s.

Tips to review "Crowd" Events

Monitor dangerous crowds on the Events or Alarms screen represented by cards, each holding the time stamp of the crowd's birth in the lower-right corner of the card and the number of people in it.

Remember, "Crowd" is a durable event, which means that once built up, a crowd may later dissipate, about which the module does not release any notification. However, whether the crowd still persists or when it has dissipated can be checked. To do so, click the event card to play back the event and take a look at the event duration indicator:

Here:

1- crowd built up
2 - crowd dissipated

The playback timeline might be delayed with respect to the time stamp of the event on the card due to the effect of the Recount every ... s parameter.

Estimate a crowd "on the fly"

Usually, viewing a playback is enough to verify a crowd event, since the Platform tests demonstrate over 95% true positive probability in foreseeable normal conditions, plus the event itself is not very sensitive to the exact number of people in a crowd. However, more accurate estimates of module performance efficiency can only be made in each particular Customer's conditions.

While a video is playing in the media player from a camera that has Count people by = Top of head with  Save people's position switched on, the ROI can be rebuilt to estimate a crowd numerically "on the fly" in the new ROI. To do so, click 1 as shown, then click Count (2) to see the result in the bottom-left corner of the screen.