How tanhuy2k2/exercise-support-system works
This is a browser-based AI exercise coaching system with a Flask API. The main workflow captures webcam frames, recognizes a returning user or collects profile details, then runs pose tracking, posture/repetition feedback, and exercise prescription. Profile and progress data live in Qdrant; prescription tables are read from Google Sheets. The README documents EMA smoothing and 33 pose landmarks. The available excerpts show the frontend worker/API calls and backend storage/sheet calls; exercise-session logic is excerpted only in part, so unsampled model and UI wiring is omitted.
Main entry points: detect_face.js Face and profile · Predict_AgeGender.js Age gender inference · main.py Flask routes · pose_estimation.js Exercise session
Components 17
Web coaching
- Face and profile
frontend/js/detect_face.js - Age gender inference
frontend/js/Predict_AgeGender.js - API worker
frontend/js/worker.js - Exercise session
frontend/js/pose_estimation.js
API and domain
- Flask routes
backend/main.py - Face embedding
backend/main.py
Profile and plans
- Face and progress store
backend/qdrant.py - Exercise prescriptions
backend/Access_sheet.py
Inference assets
- MobileNetV3 model
model/MobileNetV3L_2.onnx - Age gender labels
model/dic_labels_age.json - Gender labels
model/dic_labels_gender.json - Push-up classifier
model/MobileNet.onnx
Other components
- Trainee
- Google Sheets
- Qdrant service
- MediaPipe Pose
- MediaPipe face detect
Connections 16
- Trainee → Face and profile: provides webcam
README.md - Face and profile → MediaPipe face detect: detects face
frontend/js/detect_face.js - Face and profile → Age gender inference: predicts profile
frontend/js/detect_face.js - Age gender inference → MobileNetV3 model: runs inference
frontend/js/Predict_AgeGender.js - Age gender inference → Age gender labels: loads labels
frontend/js/Predict_AgeGender.js - Age gender inference → Gender labels: loads labels
frontend/js/Predict_AgeGender.js - Face and profile → API worker: posts requests
frontend/js/detect_face.js - Exercise session → API worker: posts requests
frontend/js/pose_estimation.js - API worker → Flask routes: calls REST routes
frontend/js/worker.js - Flask routes → Face embedding: embeds images
backend/main.py - Flask routes → Face and progress store: verifies and updates
backend/main.py - Face and progress store → Qdrant service: reads and writes
backend/qdrant.py - Flask routes → Exercise prescriptions: requests plans
backend/main.py - Exercise prescriptions → Google Sheets: reads worksheets
backend/Access_sheet.py - Exercise session → MediaPipe Pose: tracks landmarks
frontend/js/pose_estimation.js - Exercise session → Push-up classifier: uses classifier