Applying A.I. zones to live video

Zonal AI allows the user to select a specific A.I. model to apply to a zone within a live stream video. This tutorial will guide you through the steps required to apply this to your video. These are the things you need to take care of before applying for zones:-

  • Activate your live-stream video by adding a device and start the live stream
  • Activate your A.I. model from the AI app store
  • In the live stream dashboard, select the AI toggle to activate the AI model in the live stream

Note: 

  • When setting up the zones the device (camera) needs to be live streaming
  • Zones are configured per device and per A.I. model
  1. Select the Device and A.I. model
  2. Select ADD ZONES

Choose a Zone Type

There are currently two types of zones available to apply to the live video

  • Polygon - designed to identify and count objects within a designated region
  • Bi-directional - designed to count the number of people moving across a designated line in two directions

Polygon Zones

The use of polygon zones suits use cases requiring a larger area or zone that also has an unusual shape.

  • Identify objects in created zones
  • For use with various A.I. models with zones enabled
  1. Select Polygon and ADD
  1. Click on the image to create points and draw out the polygon zone
  1. Press ENTER on the keyboard when complete
  2. Make adjustments to zones by using corner points to move corners, edge points to scale edges and the white point in the middle to rotate
    1. For precise control when rotating click and hold the rotation point and move the cursor further away from the point 
  1. Click ADD to configure additional zones
    1. Click text to rename zones
    2. Use the BIN icon to delete the zone
  2. Click SAVE when finished
    1. Click X, top right, to exit without saving
    2. RESET to return to last saved point

Polygon Zone Placement

  • For reliable results, make sure that the zones do not overlap
  • Objects that are detected outside zones will be identified as “None” on the CSV output

Bidirectional Zones

Bidirectional zones are useful for use cases in which the number of movements across a given location is required to be monitored

  • Count objects that are moving across a line and track the direction of the movement
  • For use with Bi-Directional People Counter A.I. model
  1. Select Bi-directional and ADD
  1. Use the white point in the middle to rotate, edge points to scale edges and corner points to move corners.
  1. For precise control when rotating click and hold the rotation point and move the cursor further away from the point 

Note:

  • The edge the arrow is pointing to is the “EXIT” line and the opposite is the “ENTER” line. The object will be counted when a specific point of the bounding box crosses the line in the correct direction
  • The direction of the arrow points to “EXIT” and the bounding box will be counted when the object crosses the “EXIT” line in this direction.
  • The point on the bounding box can be configured and by default, the point is the middle of the bottom edge of the bounding box.
  1. Press ENTER on the keyboard after adjusting the zone 
  2. Click ADD to configure additional zones
    1. Click text to rename zones
    2. Use the BIN icon to delete the zone
    3. RESET to return to last saved point
  3. Click SAVE when finished
    1. Click X, top right, to exit without saving
    2. RESET to return to last saved point

Bi-directional Zone Placement

For accurate counts it is important to place zones:

  • where people will move through the area and also consider how they use the space 
  • where there are good and consistent detections

Need to avoid areas where people will linger in the scene and cause false positives.

  • Need to avoid situations where the person moves back and forth across the zone edges
  • If a person or object is detected standing in the line, the count can increase fairly quickly 

Example for the above image

  • It is highly likely that a person will move through the gate area  
  • Due to the camera angle, the view in the gate area is obstructed and the front area gave poor detections because of the fish eye camera. 
  • The zone was moved to inside the gates, but there is a higher chance that a person will linger in that area and cause false positives

Use of Tracklets

  • Tracklets show where the various objects are detected and tracked over the recorded footage. 
  • For the tracklets to be used effectively, the point on the bounding box where the lines are drawn needs to be the same as the Bi-directional A.I. 
  • Tracklets can help identify where to place the bidirectional zones.

Shows there are minimal detections in the gate area and also shows people lingering on the scene

There are better detections inside the gate area compared to at or outside the gates