摘要
为解决传统步态识别模型训练时间长、识别准确率低等问题,文章构建了能够进行深度学习的行走时步态特征数据库,通过对所选取特征的特定性、关联性和稳定性的研究,建立步态特征矩阵,利用Tensorflow设计CNN算法进行深度学习,使其能够自动实现基于行走时步态特征的个人识别。在27名实验者情况下,该模型的识别准确率可达99%以上,且训练时间较短,优于目前已发表的其他模型,对构建更大数据库的识别系统具有启发意义。
In order to solve the problems of long training time and low recognition accuracy of traditional gait recognition models,a walking gait feature database capable of deep learning is built.Through the study of the specificity,relevance and stability of the selected features,a gait feature matrix is established.The CNN algorithm designed by Tensorflow is used for deep learning,so that it can automatically realize personal recognition based on walking gait features.In the case of 27 experimenters,the recognition accuracy of the model can reach more than 99%,and the training time is shorter,which is superior to other published models.It is instructive for building a recognition system with a larger database.
作者
张亦鸣
王秋轶
吴梓睿
李文琳
朱鹏宇
丁浩
Zhang Yiming;Wang Qiuyi;Wu Zirui;Li Wenlin;Zhu Pengyu;Ding Hao(Jiangsu Police Institute,Nanjing 210000,China)
出处
《无线互联科技》
2023年第12期139-141,共3页
Wireless Internet Technology
基金
江苏警官学院学生科研项目课题,项目编号:202210329035Y。
