How leo4bits/poserecognition works
This is a small video action-recognition script: it reads video frames, tracks people with YOLO pose, classifies actions from keypoints and per-person history, annotates frames, and writes a video. The graph reflects the single sampled implementation; there is no README describing additional capabilities.
Main entry points: model_OOP.py Action Recognizer
Components 10
Runtime
- Script Entry
model_OOP.py - Action Recognizer
model_OOP.py - Input Video
- Track History
model_OOP.py
Pose And Actions
- YOLO Pose Tracker
- Action Classifier
model_OOP.py - Keypoint Geometry
model_OOP.py
Video Output
- OpenCV
- Annotated Video
Other components
- Operator
Connections 12
- Operator → Script Entry: starts
- Script Entry → Action Recognizer: creates and runs
- Action Recognizer → OpenCV: opens video
- Input Video → OpenCV: read by
- Action Recognizer → YOLO Pose Tracker: tracks frames
- YOLO Pose Tracker → Action Recognizer: returns detections
- Action Recognizer → Action Classifier: classifies keypoints
- Action Classifier → Keypoint Geometry: calculates distances and angles
- Action Classifier → Track History: reads and updates
- Action Classifier → OpenCV: draws keypoint lines
- Action Recognizer → OpenCV: annotates and writes frames
- OpenCV → Annotated Video: encodes frames