Open topic with navigation
Train Scene Analysis
To run scene analysis, you must create a training configuration that specifies how to detect objects and describes the events that you want to detect. A configuration describes all of the events that you want to detect from a single camera.
A configuration can include multiple categories. A category describes a single type of event for which you want to generate alarms, for example a vehicle breaching a red light or a person entering a restricted area.
Media Server includes the scene analysis training utility, a Windows application that you can use to rapidly train scene analysis. You can use the Training Utility to:
- Define regions of interest for each category in your configuration.
- Mask parts of the scene that you do not want to monitor for any category.
- Define the size, shape, orientation, velocity, and color of the objects that you want to detect, and the permitted variations in all of these properties.
- Define the position of traffic lights in the scene, so that Media Server can read the lights and generate alarms if an event occurs while the lights are red.
- Display the video being analyzed by Media Server, with an overlay that shows objects being tracked, so that you can confirm objects are tracked correctly.
- Set up filters to reduce the number of false alarms. For example, you might want Media Server to generate alarms only for objects that remain in the scene for a certain amount of time.
- Review the alarms that have been generated using your training, and classify each one as a true alarm for a specific category, or as a false alarm. The training utility can then optimize the training to minimize the number of false alarms and the number of missed alarms.