How to Use#

This guide covers daily operational use of the Take-Away Order Accuracy system.

Operational Modes#

Single Worker Mode#

Best for development, testing, and single-camera setups.

make up

Endpoints:

Gradio UI: http://localhost:7860
REST API: http://localhost:8000
MinIO Console: http://localhost:9001

Parallel Worker Mode#

Best for multi-camera deployments. Pass WORKERS to set the number of RTSP stations.

make up-parallel WORKERS=3

This starts the rtsp-streamer service and enables the frame-selector service for all stations.

Using the Gradio Web Interface#

Access the Gradio UI at http://localhost:7860.

Upload a video file or enter an RTSP URL
Enter the expected order items (one per line)
Click Validate Order
Review matched, missing, and extra items in the results panel

Using the REST API#

Base URL: http://localhost:8000

Health Check#

curl http://localhost:8000/health

Upload Video#

Uploads a video file and starts the GStreamer processing pipeline:

curl -X POST http://localhost:8000/upload-video \
  -F "file=@video.mp4"

Run From Any Source#

Start processing from a file, RTSP stream, or other video source:

curl -X POST http://localhost:8000/run-video \
  -H "Content-Type: application/json" \
  -d '{
    "source_type": "rtsp",
    "source": "rtsp://camera:554/stream"
  }'

Get Results#

curl http://localhost:8000/results/<order_id>
curl http://localhost:8000/vlm/results

Additional Endpoints#

Endpoint	Method	Description
`/mode`	GET	Current service mode and worker count
`/statistics`	GET	Processing statistics
`/videos/history`	GET	All processed videos
`/videos/summary`	GET	Video processing summary
`/videos/current`	GET	Currently processing video
`/videos/{video_id}`	GET	Specific video details
`/videos/{video_id}/complete`	POST	Mark video complete
`/videos/{video_id}/fail`	POST	Mark video failed
`/videos/history`	DELETE	Clear video history
`/orders/{order_id}/recall`	GET	Recall frames for an order
`/orders/{order_id}/replay`	GET	Replay order processing

Configuration Tuning#

Frame Selector (`config/application.yaml`)#

frame_selector:
  top_k: 3 # Frames selected per order
  min_frames_per_order: 1 # Minimum frames required before processing
  poll_interval_sec: 1.5 # MinIO polling interval (seconds)
  min_frames_before_finalize: 5 # Wait for N frames before finalizing
  inactivity_timeout_sec: 8 # Seconds of inactivity before order ends

VLM (`config/application.yaml`)#

vlm:
  temperature: 0.2 # Lower = more deterministic (recommended)
  max_new_tokens: 100 # Token limit for VLM output

RTSP Streams (parallel mode)#

Configure station streams via the RTSP_STREAMS environment variable in .env:

RTSP_STREAMS=rtsp://camera1:554/stream,rtsp://camera2:554/stream

Scaling Mode#

# Fixed worker count (default, predictable performance)
SCALING_MODE=fixed

# Auto-scaling
SCALING_MODE=auto

Monitoring and Logging#

# Check all services
make status

# Test API health
make test-api

# View logs
make logs               # order-accuracy service
make logs-vlm           # OVMS VLM service
make logs-gradio        # Gradio UI
make logs-all           # All services combined

Service Health Endpoints#

curl http://localhost:8000/health       # order-accuracy API
curl http://localhost:8001/v1/config    # OVMS VLM
curl http://localhost:9001              # MinIO Console