ShudhServer

A Gabriel cognitive engine server that performs real-time object detection on video frames streamed over websocket.

Frames are received from a Gabriel client (e.g. a wearable or mobile camera), run through a YOLOv3 model trained on the COCO dataset, and detection results are returned to the client. This is an early edge-computing / wearable cognitive assistance experiment, related to picamcommunication and yolotrial in this org.

How it works

• Loads a YOLOv3 model (yolov3.cfg + yolov3.weights, not checked into this repo) via OpenCV's DNN module
• Classifies detected objects against the standard COCO class list (coco.names)
• Runs as a Gabriel local_engine server, listening on port 9099 (see common.py)
• Logs request latency to Loglatency.log

Requirements

See requirements.txt:

• gabriel-client / gabriel-server (2.0.1)
• opencv-python
• torchvision
• PyQt5
• py-cpuinfo

You'll also need YOLOv3 weights (yolov3.weights) alongside the included yolov3.cfg, which aren't included in this repo due to file size.

Running

pip install -r requirements.txt
python server.py