Real-time ergonomic analysis with computer vision
Cameras on the production line assess operator posture and flag ergonomic risks in real time.
- Role
- Backend development: inference service and real-time gateway
- Client
- Manufacturer (client under NDA)
Challenge
Assess the posture of several operators across several cameras at once, answering fast, with risk rules the company can tune without code changes.
Solution
Message-queue-based microservices. One service captures the cameras, an inference service runs the models (YOLO in ONNX with GPU, MediaPipe pose and hands), another computes angles and risk levels from configurable rules, and a gateway streams results to the frontend in real time (video and events). A 3D skeleton view colored by risk is included, and frames can be anonymized. All in Docker.