Leak Detection Demo

Sample code for leak detection using FrameGrab and Groundlight ML services

Running the Script

You can install all the necessarily dependencies with uv or pip

uv venv
uv sync

pip install -r requirements.txt

Configuring the Script

The script requires three detectors to run:

• Binary Detector detect_leaks: Is there a leak or spill on the floor?
• Counting Detector count_leaks: Label each leak or spill in the image
• Multi-class Detector classify_leaks: What types of spill or leak is this?

In the example, we used the following classes for classify_leaks detector:

1. Water Spills
2. Residue Water After Cleanup
3. Food Residue (Including Sauces)
4. Other

After creating the detectors, modify the yaml file in configs/config.yaml to include the detector IDs. You can also configure other settings inside the yaml file.

• endpoint: Specify the edge-endpoint address, default https://api.groundlight.ai to use cloud
• detect_leaks: Binary detector ID to detect is there a leak or spill
• count_leaks: Counting detector ID to count and obtain the bounding boxes for all the leaks in the image
• classify_leaks: Multi-class detector ID to classify what types of spills or leaks
• enable_motion_detection: Enable motion detection
• motion_detection_threshold: Set the percentage of minimum pixels required to trigger a motion detection

endpoint: "https://api.groundlight.ai"
leak_detector_ids:
  detect_leaks: "det_"
  count_leaks: "det_"
  classify_leaks: "det_"
enable_motion_detection: false
motion_detection_threshold: 0.1

Configuring the Camera

The sample script uses FrameGrab to get frames from a variety of sources. The configuration file can be found and configured in configs/camera.yaml. Please refer to the FrameGrab documentation here of how to configure the camera.

After the modify the configuration files, the script can be ran with the following command:

Poetry

uv run python detect_leak.py

Pip

python detect_leak.py