Thesis Nest built a real-time action recognition system that analyzes video streams to detect activities like walking, running, fighting, falling, loitering, and unsafe behaviors—powering safety alerts and analytics for campuses, factories, hospitals, and smart cities.
Highlights:
- Spatiotemporal deep models (RGB + optional pose) for robust recognition
- Works with RTSP/ONVIF cameras; runs on edge (Jetson/IPC) or cloud
- Low-latency alerts, confidence scoring, and case review timeline
- Privacy-first: on-prem processing, event metadata only, configurable retention
- Dashboard & API: live overlays, heatmaps, and exportable incident reports
Impact: Faster incident response, fewer false alarms, and actionable behavior insights—without replacing your existing cameras.
