Runway FOD AI Detection System
Budget: $250 – $750 USD
I need an end-to-end AI solution that lets a runway-patrol drone spot even the tiniest foreign object debris (FOD) in real time. The airframe already carries LiDAR, an infrared sensor and an electro-optical (EO) camera; what is missing is the software that fuses those feeds, classifies debris and reports exact GPS position so ground crews can clear it fast.
Scope of detection
• Metallic pieces, plastic fragments and loose stones—down to roughly 1 mm in size.
• Operation must be reliable in both daylight and nighttime conditions.
• Accuracy target is set to a high threshold; false positives have to stay low while recall stays near perfect.
Key tasks
1. Build or adapt a multi-sensor fusion pipeline that consumes LiDAR point clouds, IR imagery and EO frames.
2. Train or fine-tune the detection model to differentiate FOD from background, taxiway lights, heat haze, etc.
3. Output a GPS fix (lat/long) for every confirmed object.
4. Provide a lightweight runtime that can run either on the drone’s onboard computer or a nearby edge server with minimal latency.
5. Deliver documented test results on real or simulated runway scenes demonstrating the required accuracy in day and night scenarios.
Deliverables
• Source code and trained weights (TensorFlow, PyTorch or similar).
• Deployment script or container.
• Brief technical report summarising methodology, training data used, evaluation metrics and performance.
Acceptance criteria
When I feed the system a representative data set captured from the drone, it must detect ≥95 % of the target items with <5 % false alarms and return GPS coordinates within a metre of ground truth.
If you have previous experience with FOD detection, sensor fusion, LiDAR segmentation or edge-AI optimisation, please highlight it when you respond.
Scope of detection
• Metallic pieces, plastic fragments and loose stones—down to roughly 1 mm in size.
• Operation must be reliable in both daylight and nighttime conditions.
• Accuracy target is set to a high threshold; false positives have to stay low while recall stays near perfect.
Key tasks
1. Build or adapt a multi-sensor fusion pipeline that consumes LiDAR point clouds, IR imagery and EO frames.
2. Train or fine-tune the detection model to differentiate FOD from background, taxiway lights, heat haze, etc.
3. Output a GPS fix (lat/long) for every confirmed object.
4. Provide a lightweight runtime that can run either on the drone’s onboard computer or a nearby edge server with minimal latency.
5. Deliver documented test results on real or simulated runway scenes demonstrating the required accuracy in day and night scenarios.
Deliverables
• Source code and trained weights (TensorFlow, PyTorch or similar).
• Deployment script or container.
• Brief technical report summarising methodology, training data used, evaluation metrics and performance.
Acceptance criteria
When I feed the system a representative data set captured from the drone, it must detect ≥95 % of the target items with <5 % false alarms and return GPS coordinates within a metre of ground truth.
If you have previous experience with FOD detection, sensor fusion, LiDAR segmentation or edge-AI optimisation, please highlight it when you respond.