serhannn
19th October 2015, 09:21
I would like to extract motion vector, macroblock type information and DCT coefficients from a compressed H.264 bitstream. I will then process obtained compressed domain features for an action recognition and tracking application.
What would be the most suitable tool for the job? So far, I've been using ffmpeg libraries and the provided example extract_mvs.c to extract this information. However, here I just have to rely on the accuracy on these ffmpeg functions and I can't get more detailed information (such as if some macroblocks are coded in skip mode, quarter-pel accurate motion vectors, etc). Would I be able to get more information if I modify the reference decoder (JM) or do you think it is better to use ffmpeg for this kind of job?
What would be the most suitable tool for the job? So far, I've been using ffmpeg libraries and the provided example extract_mvs.c to extract this information. However, here I just have to rely on the accuracy on these ffmpeg functions and I can't get more detailed information (such as if some macroblocks are coded in skip mode, quarter-pel accurate motion vectors, etc). Would I be able to get more information if I modify the reference decoder (JM) or do you think it is better to use ffmpeg for this kind of job?