/keypoint_correction

Primary LanguagePython

Algorithm Used

Tracking Algorithm Used

  1. Sparse Optical Flow Sparse feature point have been detected b/w two consicutive key frame cv2.calcOpticalFlowPyrLK(prev_gray, curr_gray, points_prev, None, **lk_params)

  2. Dense Optical Flow Dense feature point have been detected b/w two consicutive key frame cv2.calcOpticalFlowFarneback(prev_gray, curr_gray, None, 0.5, 5, 15, 15, 7, 1.5, 0)

Smoothening Algorithm

We can use low pass filter to smoothen the keypoints. I have implemented One Euro Filter for one of my project but due to time constraint I could not implement it here (https://gist.github.com/3846masa/5628f711e86fd62bea56b18e32177c60)

How to run the code

  1. Create & activate environment using conda conda create -n onform python=3.10 conda activate onform

  2. Install dependency pip install -r requirements.txt

  3. Run the prediction file

positional arguments: video_path Path to the input video file json_file_path Path to the JSON file containing keypoints output_video_path Path to the output video file

e.g. python predict_landmarks.py assets/test_data/test2/video.mp4 assets/test_data/test2/Old_JSON.json assets/test_data/test2/output.mp4