Compute optical flow images for VIRAT datset using successive frames, save the images while maintaining the file structure of RGB images. Verify if the any images have been corrupted during computations and recompute them.
The current implementation supports BROX and TVL1 algorithms and uses their gpu implementation in OpenCV to compute the flow vectors. Average latency of the algorithm is 4-5 fps. The images maintain the resolution of the original images.
The output images are generated from flow vectors by converting the optical flow vectors to polar coordinates and using the magnitude and angle to populate the hue and value plane of an hsv image. The hsv image is converted to rgb image and saved on the disk.
Compute optical flow images for VIRAT datset using successive frames, save the images while maintaining the file structure of RGB images. Verify if the any images have been corrupted during computations and recompute them.
The current implementation supports BROX and TVL1 algorithms and uses their gpu implementation in OpenCV to compute the flow vectors. Average latency of the algorithm is 4-5 fps. The images maintain the resolution of the original images.
The output images are generated from flow vectors by converting the optical flow vectors to polar coordinates and using the magnitude and angle to populate the hue and value plane of an hsv image. The hsv image is converted to rgb image and saved on the disk.