TRASSIR Pose Detector or simply Pose Detector is designed for building video surveillance systems that require detailed image analysis using neural networks. The detector uses algorithms for analyzing human movement and behavior to determine a person's pose in the frame. This allows video surveillance operators to detect atypical or suspicious behavior in the surveillance zone in real time, such as falling or raising their hands during an attack.
The detector can recognize the following postures of a person:
- sitting, including leaning on knees or elbows after a fall;
- bent;
- lying flat with arms and legs relaxed and extended;
- with one or both arms raised and elbows raised (if elbows are low, detection may fail).
All other poses are classified by the detector as regular.
The pose detector flexibly identifies various human behavior scenarios. For example, a person resting on their hands and knees is classified as "sitting," while a "lying" pose is only recorded when the body is completely relaxed and lying flat. The detector also reliably recognizes a "bent" pose. This enables an accurate response to a fall, which can be recorded as one of three poses: "sitting," "bent," or "lying."
The detector is integrated with a notification system that promptly sends information about recorded events.
Setting up the Pose Detector is performed in three steps:
- Setting up an analytics server for the pose detector - necessary for analyzing video frames.
- Configure Pose Detector for a channel - video from the channel will be used to detect human poses.
- Setting up a dashboard for the pose detector - necessary for the analysis of the detector's operation.
