Components

OMNI Agent captures video streams from cameras and video files using RTCP protocols. The received frames are processed in parallel within face, body and skeleton processing modules.

OMNI Agent modules detect faces in the frames, identify facial attributes, build face biometric templates, and search for face matches wth the database.

Simultaneously, the modules perform face / body / skeleton tracking: from frame to frame, the displacement vector of the face / body /skeleton area is calculated, confirming or refuting the prediction of the next face/body/skeleton coordinate with false-positive filtering.

Generated tracks of faces, bodies, and skeleton joints are used as input data for detecting the presence of people in a region of interest (ROI) and for identifying crossings of virtual lines in the video stream / video file. Body tracks are also utilized for human action recognition (HAR), including actions such as fights, falls, lying, and sitting, as well as for tracking people's movement on a room map (Map tracking).

The collected data for each detected person, formatted as processes and events, is transmitted to OMNI Platform or external service (via webhook / MQTT) for further processing.