Y-B-Class-Projects/Human-Fall-Detection — reverse-engineered prompt

Build me a Python project that can detect when a person falls, both from a saved video file and from a live webcam feed. I want it to use the existing pose based approach in this repo, where each person gets 17 body keypoints, then a simple rule based fall detector decides if the person has likely fallen based on body position and motion over a short time window.

Please make the whole thing runnable end to end, including loading the pose model weights, processing frames, drawing the keypoints and a clear fall or no fall label on screen, and saving or showing the output video. Keep the fall logic configurable with environment settings for things like fps, window size, movement threshold, vertical drop, and body aspect ratio.

If the older learning based model is still in here, leave it as optional, but make the rule based path the main one since that seems to be the approach that actually works. Also clean up anything needed so someone can install requirements and run video mode or real time mode without guesswork. Look up current docs online if you need to.