Skip to main content
Version: 2.6.0

Architecture

Basic ModulesDescription
Activation ModuleThe module is responsible for Agent activation.
Camera HandlerThe module splits video streams from cameras or video files into frames and prepares them for processing.
Human Process HandlerThe results of processing human faces and bodies are submitted to the Human Process Handler, where they are converted into agent processes and events for subsequent transfer to OMNI Platform or an external service via webhooks/MQTT.
Body Frame Processing ModulesDescription
Body Detector Processing BlockThe module detects human bodies in frames.
Byte Track Processing BlockThe module implements the Byte Track algorithm to improve the detector quality.
View Transform HandlerThe module transfers the coordinates of people detections from the camera to the map.
Multistream Mapping HandlerThe module combines the position of detections on the map from different cameras.
Embeddings Processing BlockThe module creates embeddings from body crops. Embedding means a vector representation of the detected body.
Reidentification HandlerThe module forms a person's track and provides reidentification. During the detection process, each frame is assigned a unique track_id. If the same person is identified on a sequence of frames, the track_id of all related frames are changed to the track_id of the first frame of this person.
Body Track HandlerThe module organizes and processes the ways people move on frames (tracking).
Hpe Resnet V1D Processing BlockAn internal component of HAR pipeline that determines the position of the human skeleton on the frame.
Har Container Processing BlockThe module aggregates the positions of the human skeleton in time into a common container.
Har Pose C3D Processing BlockAn internal component of HAR pipeline that specifies the class of person's position.
Har Hll Processing BlockThe module detects an action.
Face Frame Processing ModulesDescription
Videoworker Callbacks HandlerThe module detects human faces.
Face Track HandlerThe module estimates face attributes in the frame (gender, age, emotions, liveness, mask presence), searches for similar faces in the agent and platform databases (via the Database Sync Handler). If there are matches with the database, the ID of the image from the database is also added to the processing result.
Attentions HandlerThe module estimates human attention (if the person in the frame is looking at the camera).
Infrastructure ModulesDescription
Webhooks HandlerThe module sends Agent processes via webhooks to an external service.
Server Connection HandlerThe module transfers Agent processes to OMNI Platform.
Webhook Data Projection HandlerThe module that projects data into a JSON-like document for subsequent transmission via webhooks.
Server Data Projection HandlerThe module that projects data into a JSON-like document for subsequent transfer to the OMNI Platform.
Database Sync HandlerThe module synchronizes the local database with OMNI Platform database.
Reconfiguration HandlerThe module provides access to configuration files and allows to reconfigure agent without app restart.
Preview HandlerThe module creates previews (frames from the camera with rendered detections, submitted to the web interface of OMNI Agent).
Frontend HandlerThe module manages the agent as a web app.
External Process HandlerThe module brings the signals of the beginning and end of external events to an interval form (processes).
Http Redirection ModuleThe module redirects requests to the platform server.
Multisource HandlerThe module creates a pipeline of processing modules for each new videostream.
Trace HandlerThe module collects traces (logs) from Body Detector.
Custom Resources ProviderThe module provides information about the number of visible people and frames from the camera through the internal API.
MQTT Data Projection HandlerThe module projects process data for subsequent sending via a local MQTT queue.
MQTT HandlerThe module sends data to a local queue.