Bug fix vector_frames size in optical_flow.py

[FMarmoreo]

Bug fix vector_frames size in optical_flow.py since the last frame of zeros in the initialization of vector_frames is not filled by the horn-schunck algorithm that runs in range frames[:-1]. This, in particular, introduces errors also in smooth_frames function.

[cosimolupo]

@rgutzen why is it important for vector_frames to have the same shape as frames? Is this feature explicitly or implicitly used somewhere downstream in the code?