Extracting Discriminative Features for Cross-View Gait Recognition Based on the Attention Mechanism

Human identification based on gait biometrics has become a popular research topic of computer vision and pattern recognition due to its great potential in public security and surveillance system. However, the recognition accuracy can be seriously degraded because of the appearance differences caused by view angle variation. To tackle this problem, we propose a method based on convolutional neural network (CNN) and attention mechanism to solve the cross-view problem in gait recognition. In the proposed algorithm, we firstly extract the features based on CNN structure and then the Horizontal Splitting operation is done to obtain the feature partitions in different granularities. After that, the attention mechanism is utilized to calculate the attention scores of the input partitions on both spatial and channel domain and finally the group of feature vectors can be obtained to determine the corresponding identity. In order to verify the effectiveness of the proposed method, the experiments are done based on two popular gait datasets–CASIA-B and OU-ISIR LP. The results show that the proposed model can effectively extract the discriminative gait features robust to view angle variation and improve the crossview gait recognition accuracy compared with the state-of-the-arts.