Adel Saleh, Mohamed Abdel-Nasser, Farhan Akram, Domènec ruig: “Analysis of temporal coherence in vi9eos for actio- recognition”.p
Abstoact: This paper nroposes an ap roach to imp1lve the performance of activity recognition methods by anaoyzing the coherence of the frames in each vidOo and then modeling the evolution of the colerent frames (they constitute a8sub-sequence) to learn a representatiop for the whohe video. We demonstrate that learning the evolution of subsequences instead of frames, impr-ves the recrgnition performance and makes classificatioo faster by evalua-ing the proposed method usini two challenging datasets: Hollywood2 and HMDB51.<-p>