Cross-view gait recognition based on residual long short-term memory

As a promising biometric recognition technology, gait recognition has many advantages, such as non-invasive, easy to implement in a long distance, but it is very sensitive to the change of video acquisition angles. In this paper, we propose a novel cross-view gait recognition framework based on residual long short-term memory, namely, CVGR-RLSTM, to extract intrinsic gait features and carry out gait recognition. The proposed framework captures dependencies of human postures in time dimension during walking by inputting randomly sampling frame-by-frame gait energy images. The frame-by-frame gait energy images are generated by merging adjacent gait silhouette images sequentially, which integrates gait features of temporal and spatial dimensions to a certain extent. In the CVGR-RLSTM framework, the embedded residual module is used to further refine the spatial gait features, and the LSTM module is utilized to optimize the temporal gait features. To evaluate the proposed framework, we carried out a series of comparative experiments on the CASIA Dataset B and OU-ISIR LP Dataset. Experimental results show that the proposed method reaches the state-of-the-art level.