mav-gate-detection

Individual assignment for course Autonomous Flight of Micro Air Vehicles 2020/2021

Dependencies

Requirements: OpenCV, PyTorch, Jupyter Notebook

Run

Put the dataset WashingtonOBRace under the repo. Download saved models: https://drive.google.com/drive/folders/1_L6Vzn01lf_hUjFGckGz6qNewVFDrCHq?usp=sharing. Then put them in the folder learned_weights/ in the repo root directory.

The codes is with the same structure of the report. For only checking the proposed method(report section 3), just run gate_detection.ipynb.

• Report section 2.1

  • gate_detection_orb.ipynb : explore the ORB features extraction method(if it works, filter good matches, visualize, etc)
  • gate_detection_orb_test.ipynb: finally test the ORB features method(computational effort, ROC curve)

• Report section 2.2

  • gate_detection_cnn.ipynb: explore the binary classification method. Skip the node for training the model, and later in the code 'learned_weights/gate_det_cnn.pt' is loaded for testing. Actually for avoiding accidentally starting training I have commented them.

• Report section 2.3

  • gate_detection_maskrcnn.ipynb: explore the Mask R-CNN method. Skip the node for training the model, and later in the code 'learned_weights/gate_det_maskrcnn' is loaded for testing

• Report section 3

  • gate_detection.ipynb: test the final proposed method. Run the notebook step by step