AI-Based Ergonomic analysis for Workers in Warehousing

We present a lightweight, privacy-preserving ergonomic assessment tool for warehousing that runs entirely client-side in a web browser. From ordinary images or video frames, the system estimates full-body pose landmarks, computes interpretable joint angles (e.g., trunk, neck, shoulder, elbow, knee, wrist), and maps them through the published REBA/RULA rule tables to produce an overall risk score and action level. Users can optionally provide task context (e.g., task type, repetition rate, coupling, load/force proxies) to improve scoring consistency across common warehouse activities, including lift/lower, carry, push/pull, and pick/place. Because computation remains on-device, no video or keypoints are uploaded, enabling practical field deployment with reduced privacy concerns. We emphasize transparency by exposing the intermediate table lookups and thresholds used to reach each score. We evaluate the approach using (i) joint-angle accuracy against a skeleton-based reference dataset and (ii) qualitative demonstrations in realistic warehouse scenes, and we report interactive performance on commodity devices. Limitations include occlusion, extreme camera viewpoints, and unknown external loads unless provided by the user.