AI-Powered Factory Safety Detection

The system detects four violation types: (1) No helmet: workers without head protection; (2) No gloves: workers with bare hands where hand protection is required; (3) Smoke/lighter/matchstick usage: fire-risk behavioral violation; (4) Mobile phone usage: distraction-related behavioral violation. All four run simultaneously in real-time.

No. The system runs entirely on a standard computer using the built-in or USB webcam for live detection, and any MP4/AVI video file for batch processing. No Jetson devices, no factory CCTV integration, and no IoT hardware are needed at this stage.

Helmet detection achieves >82% mAP@50: the most well-studied PPE detection type with proven research results. Gloves and behavioral detections (phone, smoke) achieve 70–78% mAP@50. Confidence thresholds are configurable per violation type (0.65–0.70 default). All detections include a per-incident confidence score.

Yes. The system has two input modes: live webcam detection for real-time monitoring, and video upload detection that processes any uploaded MP4 or AVI file frame-by-frame, logs all detected violations with timestamps and screenshots, and presents them on the incident dashboard.

Live violations are flagged within 1–2 seconds of the event occurring. Each incident record includes the violation type, timestamp, confidence score, and a screenshot of the frame where the violation was detected.

Yes. CCTV integration via RTSP streams, Jetson edge devices, WhatsApp/email alerting, multi-user access, and cloud deployment are all planned for the next phase. The current version is a focused demonstration product to validate the core detection engine before scaling to factory infrastructure.