Parrot Anafi Autonomous Computer Vision Project

CV project utilising the Parrot Anafi drone for automatically detecting and tracking footballs to generate video footage.

• An Edge TPU MobileSSDV2 model was trained using google-coral. This model allows for very fast inference times (<20ms).
• Addtional models such as YOLO-4 and quantized YOLO are also being experimented with, although these have relatively slowere inference times but high detection performance.

Requirements

• Software
  • Python >= 3.6
  • Parrot Olympe SDK
  • Tensorflow lite
  • Open CV 4
• Hardware
  • Google Coral Edge TPU USB Accelerator
  • Parrot Anafi

TODO

• [X] Access video from Anafi drone via SDK
• [X] Add capability to always show video stream regardless of detection status
• [X] COllect data, train TFlite Detection model and evaluate performance
• [X] Implement Deep Sort Tracking Class to be used in conjunction with the TFLite models
• [X] Test additional detection models
• [ ] Implement and test drone controller class e.g (take-off, flight, landing etc) using Olympe