摘要
针对步态识别过程易受拍摄视角、外观变化等因素影响问题,提出一种融合点云步态模型与深度学习的步态识别算法。算法通过轻量级特征描述符(lightweight feature descriptor,LFD)提取图像特征,并将其进行特征配准;基于几何-匹配核预处理增强识别技术(gait model-key point recognition and extraction,GM-KPRE)提取人体关键点信息,在支持向量机算法中引入径向基函数核进行步态分类和识别;在公开数据集CASIA-B和Market-1501-v15.09.15上进行实验验证,实验结果表明,算法能有效提高步态识别准确率和效率。
Aiming at the challenges that gait recognition is susceptible to factors like viewpoint variations and appearance change,we propose a gait recognition algorithm that combines point cloud gait modeling with deep learning techniques.The algorithm employs a lightweight feature descriptor(LFD)to extract and align image features.It utilizes geometry-matching kernel preprocessing enhanced recognition(GM-KPRE)to extract human body key points and enhance recognition.Additionally,a radial basis function kernel is incorporated into the support vector machine for gait classification.Experimental evaluations on standard datasets CASIA-B and Market-1501(v15.09.15)demonstrate significant improvements in both accuracy and computational efficiency.Experimental validation is carried out on the public datasets CASIA-B and Market-1501-v15.09.15,and the experimental results show that the algorithm can effectively improve the accuracy and efficiency of gait recognition.
作者
关迪元
张鑫宇
王增烨
霍焱
GUAN Di-yuan;ZHANG Xin-yu;WANG Zeng-ye;HUO Yan(School of Information Engineering,Shenyang University,Shenyang 110044,China;School of Information,Liaoning University,Shenyang 110036,China;Air Traffic Control Engineering Construction Headquarters,Civil Aviation Administration of China Northeast Region Air Traffic Management Bureau,Shenyang 110043,China)
出处
《计算机工程与设计》
北大核心
2025年第8期2312-2319,共8页
Computer Engineering and Design
基金
辽宁省教育厅高校基本科研基金项目(JYTMS20231165)。
关键词
深度学习
点云步态模型
特征提取
特征配准
人体关键点
径向基函数核
步态识别
deep learning
point cloud gait model
feature extraction
feature alignment
key points of the human body
radial basis function kernel
gait recognition
