NMF based image sequence analysis and its application in gait recognition

Human identification is an important part of the intelligence system, and among them, gait recognition is more suitable in pervasive intelligence, due to its capability in identifying with low resolution images captured at a distance without subject cooperation. Existing human gait recognition methods are either too simple that cannot cope with complex scenes, or too complex to provide service at the right time and right where. In this paper, we propose a simple but efficient method for gait recognition. It not only has a high recognition accuracy, but also has a strong adaptability to different conditions, which means it can better adapt to the real environment. After data pre-processing, the method leverages nonnegative matrix factorization (NMF) to pick up high order image features. Then, it sends frames in the order of time sequence to a long short term memory (LSTM) neural network for performing classification. Experimental results show that this method can achieve high recognition accuracy, and has a strong adaptability in cross-view and multi-clothes conditions.