FallSense: YOLO pose & ByteTrack based fall detection with explainable events
It tackles the need for reliable fall detection in video streams, especially for elder care and safety monitoring. By combining YOLO26-pose for human keypoint extraction with ByteTrack for persistent tracking, it runs a finite‑state machine to label UPRIGHT, FALLING, FALLEN, and ALARM states, outputting clear event logs and rule explanations. The pipeline is reproducible, works on CPU after a single GPU inference, and includes evaluation tools for event‑level metrics. Designed for researchers and practitioners needing an off‑the‑shelf, transparent fall‑monitoring solution.
View on GitHub →kuotunyu/fall-detection-pose