关键词: 步态识别, 跨视角, 生成对抗网络（GAN）, 跳跃连接, 卷积块注意力模块

Abstract: To address the problem of decreased recognition accuracy in gait recognition caused by viewpoint variations, this paper proposes a recognition method based on a generative adversarial network (GAN) with skip connections and a convolutional block attention module (SC-GAN). First, skip connections are introduced into the generative ad鄄 versarial network to fuse feature maps in the channel domain, thereby preserving more gait-related information. Second, the convolutional block attention module is introduced to focus on more feature locations and content specific to particular viewpoints. Finally, cross-view gait recognition experiments are conducted on the gait dataset B of the Institute of Automation, Chinese Academy of Sciences (CASIA-B dataset). Experimental results demon鄄 strate that the proposed model maintains good robustness and effectiveness under multiple covariate factors.

Key words: gait recognition, cross-view, generative adversarial network (GAN), skip connections, convolutional block atten－ tion module