Continuous action segmentation and recognition using hybrid convolutional neural network‐hidden Markov model model

• DOI: 10.1049/iet-cvi.2015.0408
• Publication Date: 2016-02-19T12:58:53Z
• Authors: Jun Lei; Guohui Li; Jun Zhang; Qiang Guo; Dan Tu
• Keywords: Viterbi algorithm

Continuous action recognition in video is more complicated compared with traditional isolated recognition. Besides the high variability of postures and appearances each action, complex temporal dynamics continuous makes this problem challenging. In study, authors propose a hierarchical framework combining convolutional neural network (CNN) hidden Markov model (HMM), which recognises segments actions simultaneously. The utilise CNN's powerful capacity learning level features directly from raw data, use it to extract effective robust features. HMM used statistical dependences over adjacent sub‐actions infer sequences. order combine advantages these two models, hybrid architecture CNN‐HMM built. Gaussian mixture replaced by CNN emission distribution HMM. trained using embedded Viterbi algorithm, data train are labelled forced alignment. test their method on public dataset Weizmann KTH. Experimental results show that authors’ achieves improved segmentation accuracy several other methods. superior property learnt also illustrated.