The embedded analytics modules realize functionalities for automatic video content analysis. The modules are available as software on the AVC product family or as library for TI-DSP family (DM642; DM648).

Meta-data (tracks, bounding boxes) are available on the network in XML-format via a RTP data stream.

The following application examples are calculated on the AVC family live and on-board. The algorithms need less than 15% of CPU load. Therefore video content analysis and streaming can be done in parallel.

Tracking of Multiple Objects

Multiple objects within the image are tracked independently. New objects are denoted with a single ID (red). Object movements are tracked and denoted with a trajectory. Occlusions are resolved using movement models such that objects with opposing movements can be tracked correctly. With 15% of the calculation power of a TI DM642 up to 50% can be tracked successfully. View Video

Roundabout Analysis

The statistical measurement of traffic scenarios can be shown in the application case of a roundabout. Counting lines and regions of interest can be defined to measure specific traffic parameter (e.g. “Number of vehicles from point A to point B”). The demo video shows interrupted trajectories, which denotes hypotheses of the trajectories that are not counted. View Video

PTZ Tracking

A PTZ camera can be automatically controlled using an overview camera for detailed object tracking. The system works autonomous using a single 2-channel encoder (AVC-1) – therefore no interaction with a PC (or similar) is necessary. The demo video shows the overview camera (left) and the detailed camera view of the PTZ (right). View Video

Forbidden Area

Surveyed areas can be defined easily. The demo video shows the detection of unauthorized access even in difficult conditions (small person, moving vegetation). The region of interest turns red in case of a detected event. View Video

Movement in Forbidden Direction

The analysis of the trajectories allows the detection of objects moving in a forbidden/undesired direction. This can be evident from the demo video. Trajectories are correct and stable and can trigger an event that can be defined using a rule-based system. View Video

Abandoned Objects

The detection of abandoned/removed objects works via the detection of static changes to the background. Left items (bags or other objects) can be found using configurable detection times and sensitivity. View Video