基于生成对抗网络的步态特征提取与识别  

Gait feature extraction and recognition based on generative adversarial networks

作　　者：文佳敏 沈永良[1] WEN Jia-Min;SHEN Yong-Liang(College of Electronic Engineering,Heilongjiang University,Harbin 150080,China)

机构地区：[1]黑龙江大学电子工程学院,哈尔滨150080

出　　处：《黑龙江大学工程学报》2021年第2期60-65,共6页Journal of Engineering of Heilongjiang University

基　　金：国家自然科学基金项目(61503127)。

摘　　要：提出了一种基于生成对抗网络的步态特征提取与识别算法。以CASIA-B数据集的步态图像为研究对象,对不同视角、不同衣着或手持物的GEI特征进行分析。在54°,90°,126°3种典型角度下对网络模型的性能进行分析,发现模型的识别性能随着视角跨度增大而逐渐降低,当视角为90°时网络模型达到最高的识别准确率。在CASIA-B数据集上的实验结果表明,使用生成对抗网络实现步态识别是可行的。A gait feature extraction and recognition algorithm based on generative confrontation networks was proposed.Taking the gait images of CASIA-B dataset as the research object,and analyzed the GEI characteristics of different viewing angles,different clothes or hand-held objects.After analyzing the performance under three typical angles(54°,90°,126°),it was found that the recognition performance of the model gradually decreases as the viewing angle span increases.When the viewing angle was 90°,the network model could obtain the highest recognition accuracy.The experimental results on the CASIA-B dataset showed that it was feasible to use generative adversarial generative networks to realize gait recognition.

关 键 词：步态识别 特征提取 生成对抗网络 真/假判别器 语义判别器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]