Exploring Gait Recognition Using Dual-Stream Convolutional Neural Networks and Feature Fusion

With the rapid development of the information society, people's safety in the public environment has received more and more attention. Among the many biometric recognition technologies, gait recognition has attracted more and more attention from scholars. This paper conducted a series of studies on gait recognition methods based on convolutional neural networks. Addressing the issue of insufficient dynamic information in commonly used gait templates for recognition, we proposed a gait recognition approach using a dual-stream convolutional neural network based on spatiotemporal feature fusion. We utilized the designed feature fusion module to merge two types of features, resulting in gait characteristics. Additionally, inspired by the Bhattacharyya distance, we introduced a novel loss function called cluster loss. Combined with the Softmax loss for the recognition task, this approach not only reduced intra-class variations but also amplified inter-class differences. We validate the proposed method by simulation experiments and demonstrate high recognition rates on large-scale datasets CASIA-B and OU-MVLP, thus confirming the feasibility of the algorithm.