Argus
A real-time surveillance engine built on a deep-learning neural network: it detects and tracks every person in a scene, scores their motion frame-to-frame through an agentic perception layer, groups them spatially, and streams the signals live, at 40+ FPS, in full light or total darkness.
Overview
At its core is a deep neural network. A YOLO11 pose-estimation model detects every person in frame, and a tracking layer assigns each a persistent identity it holds across frames, so the system reasons about individuals, not pixels, and never loses the thread as people move through a scene.
An agentic perception layer scores each subject continuously for motion and stillness, with a regional body breakdown (head, torso, arms, legs) and distance-independent sensitivity, a person at the far end of a room registers the same as one up close. A micro-motion stage catches movement the skeleton misses, and a seismograph-style anomaly signal decides what's worth an operator's attention.
Spatial clustering groups people into crowds in 3D, and every signal streams live to a control surface at sub-200ms latency. Argus runs its neural inference on the edge at 40+ frames per second and operates in complete darkness via infrared depth sensing, built for surveillance, safety, and situational awareness at enterprise grade.
Capabilities
- Deep-learning neural pose estimation with persistent multi-target tracking
- Agentic perception layer scoring motion & stillness per subject, by body region
- Distance-independent sensitivity, consistent from near field to 4m+
- 3D proximity clustering for crowd and grouping analysis
- Operates in complete darkness via infrared depth sensing
- Real-time neural inference on the edge, 40+ FPS, sub-200ms, live-streamed
Next project