How tgcabuang/guardian-vision works
Guardian Vision is an Electron-based surveillance application for live, multi-camera activity recognition and post-investigation review. Its documented inference path combines person detection, pose estimation and skeleton-based action recognition; the backend also handles alerts and recording/timeline functions. A separate model-development path prepares NTU video data and trains CTR-GCN weights for inference. The excerpts show backend component imports and recorder setup, but omit most routes and inference calls, so uncertain frontend-to-backend links are marked unverified rather than asserted.
Main entry points: server_ws.py API and WebSocket · detection_core.py Detection Runner · main.js Electron Shell · training_with_cctv_aug.py CTR-GCN Training
Components 20
Desktop Application
- Electron Shell
main.js - Login Screen
login.html - Live Dashboard
ws_dashboard.js - Admin Users
admin_users.html
Live Inference
- API and WebSocket
backend/server_ws.py - Detection Runner
backend/detection_core.py - CTR-GCN Model
backend/ctrgcn_model.py - RTMPose
backend/assets - YOLOv8
backend/assets
Recording and Review
- Continuous Recorder
backend/continuous_recorder.py - Timeline Utilities
backend/timeline_utils.py - Event Clips
backend/clip_recorder.py - Video Records
Model Development
- Dataset Builder
builder and trainer/dataset_builder_combine_recommended.py - CTR-GCN Training
builder and trainer/training_with_cctv_aug.py - Training Model
builder and trainer/ctrgcn_model.py - Model Weights
backend/assets
Other components
- Operator
- Camera Streams
- NTU RGB+D Videos
Connections 22
- Operator → Electron Shell: uses
README.md - Electron Shell → Login Screen: loads first
main.js - Operator → Live Dashboard: monitors
README.md - Operator → Admin Users: manages users
- Electron Shell → API and WebSocket: starts backend
README.md - Live Dashboard → API and WebSocket: opens WebSocket
- Camera Streams → Detection Runner: supply video
- API and WebSocket → Detection Runner: imports
backend/server_ws.py - Detection Runner → CTR-GCN Model: uses model
backend/detection_core.py - Detection Runner → RTMPose: estimates pose
- Detection Runner → YOLOv8: detects people
- API and WebSocket → Continuous Recorder: initializes manager
backend/server_ws.py - API and WebSocket → Timeline Utilities: imports
backend/server_ws.py - API and WebSocket → Timeline Utilities: inserts events
backend/server_ws.py - Continuous Recorder → Video Records: writes segments
- Event Clips → Video Records: writes event clips
- Timeline Utilities → Video Records: indexes segments
- NTU RGB+D Videos → Dataset Builder: provides videos
README.md - Dataset Builder → CTR-GCN Training: creates training data
README.md - CTR-GCN Training → Training Model: trains
- CTR-GCN Training → Model Weights: generates weights
README.md - Model Weights → CTR-GCN Model: loaded for inference
README.md