khushboo-agarwal/optical-flow

Implement Lucas-Kanade optical flow estimation, and test it for the two-frame data sets provided in Python from scratch

optical-flow's Issues

image gradient using signal.convolve2d

Hi Agarwal, it seems that the kernel used in signal.convolve2d is flipped upside down and left to right. so when calculating image gradient Ix and Iy, you are getting the negative of real Ix and Iy (not It since its kernel is same when flipped), which in term resulted in wrong direction of u.

    # First Derivative in X direction
Ix = signal.convolve2d(I1_smooth,[[-0.25,0.25],[-0.25,0.25]],'same') + signal.convolve2d(I2_smooth,[[-0.25,0.25],[-0.25,0.25]],'same')
# First Derivative in Y direction
Iy = signal.convolve2d(I1_smooth,[[-0.25,-0.25],[0.25,0.25]],'same') + signal.convolve2d(I2_smooth,[[-0.25,-0.25],[0.25,0.25]],'same')