Deep-Learning Camera Fault Detection

Detecting and removing camera noise and environmental artifacts during drone operation.

Problem

Cameras on a flying drone are exposed to noise and environmental artifacts that degrade downstream perception. This project detected such faults during operation and cleaned the affected frames.

Contribution

  • Detected camera noise with OpenCV-based analysis.
  • Fed flagged frames to a deep-learning model that removes the noise.
  • Started from open-source detection models, then moved to a zero-shot detection approach to fit memory constraints.

ROS2 · OpenCV · PyTorch · Python