Object Class Recognition for Surveillance

Object class recognition can be used to identify and track objects as they move through a scene, providing a starting point to accomplish the same goals as scene analysis.

Object class recognition identifies objects using neural network technology. This simplifies configuration because you choose the classes of objects to track, such as "cars" or "people". In comparison, scene analysis requires you to specify object characteristics such as shape, size, speed, direction, or color. Object class recognition also uses the extra class information to improve tracking, for example when one object moves in front of another.

OpenText provides a pre-trained recognizer that can identify and track people, bicycles, motorcycles, cars, trucks, and buses. Although object class recognition usually requires a visual license, you can use this recognizer with a surveillance license.

Scene analysis can generate alerts for many scenarios, such as people entering restricted areas or vehicles driving through red lights. To generate the same alerts, you can use the output from object class recognition with other Media Server engines, such as:

  • AlertPath - Generates an alert when an object follows a specified path through the scene.
  • AlertRegion - Generates an alert when an object is present within a region for some amount of time.
  • AlertStationary - Generates an alert when an object is stationary for some amount of time.
  • AlertTripwires - Generates an alert when an object crosses a tripwire.
  • Count - Counts the number of objects that are present within the scene or a region of the scene.
  • TrafficLight - Determines the state of traffic lights, so that you can detect vehicles failing to stop for a red light.

The Media Server graphical user interface, available through action=GraphicalUserInterface, can help you build surveillance session configurations that are based on object class recognition.

NOTE: Before building a surveillance session configuration in the GUI, import the pre-trained surveillance recognizer into your training database. The recognizer must be given the name surveillance. For more information about importing recognizers, see Import a Recognizer.

To build a surveillance session configuration

  1. In a web browser, run the GraphicalUserInterface action, for example:

    http://mediaserver:14000/action=GraphicalUserInterface

    where mediaserver is the machine hosting Media Server and 14000 is the Media Server ACI port.

  2. Click Surveillance Configuration.
  3. Use the controls above the media player to ingest sample video from a file or stream.
  4. Capture a scene image so that you can draw regions, tripwires, and paths. To capture a still image of the scene, click Capture a still image . If you want to save the captured image with the configuration, click Set as default image. To resume viewing the live video and capture a new scene image, click Return.
  5. Use the controls to the left of the media player to build the configuration (configure alerts, count objects, and change detection options).
  6. When you have finished building the configuration, click Export to save the finished configuration and associated files to disk. In the exported zip file is a file named README.txt that explains the function of each file.