autodrive

A python implementation for Advanced Driving Assistant. (Now with GPU Support) [Watch Demo]

About the model

This model can: * Detect lane lines and fit a curve on them with a polynomial mask. * Detect objects like: cars, buses, persons, traffic lights, etc

How to run the model?

• Firstly, clone the repository and change the directory to it.
• Run the following command inside

  autodrive

  directory in the terminal/command prompt.

  python main.py
  

  NOTE: This is a default run type for this model which uses the input video stored in test_videos folder.

To test the algorithm on your own video type the following command:

python main.py --input {path to the video}

And, if you want to store the model's output on the video then specify the output video name as:

python main.py --output {a file name.mkv}

NOTE: The output video file's extension must be

.mkv

New Features

• Added GPU support to run object detection using cuda support for getting higher FPS

  Usage:

  python main.py -g  # -g for GPU
  

  NOTE: The -g inclusion while running the model will allow the model to use GPU for faster inference and higher FPS

• Added support for using TINY YOLO Model to perform even more faster detection with smaller model.

  Usage:

  python main.py -t  # -t for TINY
  

  NOTE: The -t inclusion while running the model will tell the model to use TINY Yolo model for running object detection.

Extra features of the model

You can also make changes inside the code to add more features in the models' output.Like:

• To add labels in the top of every bounding box, pass a argument named

  labels=True

  in pipeline function of main.py

  if not lane_only:
          self.lane_mask_final = self.object_detector.draw_boxes(self.frame, self.lane_mask_final, labels=True)
  

  • If you only want to mask and detect the lanes and dont want to detect objects, you can make

    lane_only

    argument in main.py to

    True

 if __name__=='__main__':
   ...
    #run the pipeline on specified video
    pipeline.run_video(input_video, output_video, lane_only=True)
   ...

Authors

• Amit Kharel - Overall works - ajeetkharel

License

This project is licensed under the MIT License - see the LICENSE.md file for details