I've been following the VideoCapture tutorial http://docs.opencv.org/3.2.0/d5/dc4/tutorial_video_input_psnr_ssim.html and it strikes me that only fully decoded frames are analysed.
Is there a way to get hold of lower level information, i.e. partial decodings of an mp4 file? Since this is all I need for my analysis, I hope it would be more efficient than decoding the entire video:
- key frames (potentially in their encoded form, but decoded is also fine)
- p-frame motion vectors... I know h264 gets a bit more complex with full inter-frame but just forward motion estimation is fine.
- the fourier space from AAC (or anything that would resemble the DTFT... I realise this is a bit more ill-defined, it's not as important)
thanks!