基于多级分层融合网络的步态识别方法  

Method for Gait Recognition Based on Multi-level Hierarchical Fusion Network

作　　者：李哲 刘勇 刘中华 欧卫华 LI Zhe;LIU Yong;LIU Zhonghua;OU Weihua(College of Information Engineerine,Henan University of Science and Technology,Luoyang 471023,China;College of Information Engineerine,Zhejiang Ocean University,Zhoushan 316022,China;College of Big Data and Computer science,Guizhou Normal University,Guiyang 550025,China)

机构地区：[1]河南科技大学信息工程学院,河南洛阳471023 [2]浙江海洋大学信息工程学院,浙江舟山316022 [3]贵州师范大学大数据与计算机科学学院,贵州贵阳550025

出　　处：《河南科技大学学报(自然科学版)》2025年第2期42-47,94,M0004,M0005,共9页Journal of Henan University of Science And Technology:Natural Science

基　　金：国家自然科学基金项目(U1504610,61962010,62262005)。

摘　　要：步态识别是一种通过个体独特的行走方式来识别个体的生物识别技术,适合无约束的环境,具有广泛的应用前景。虽然当前的方法侧重于利用基于身体部位的表示,但它们经常忽略局部运动模式之间的层次依赖性。提出了一种多级分层融合模型,用于从粗到细提取步态特征,同时所提框架整合了多级分层特征提取与不均匀分层特征提取2种策略,实现了对局部特征的细粒度提取,同时强调了局部特征间的相互关联。在广泛认可的数据集上,经过大量实验验证,所提出的方法被证实是有效的。该方法在提升模型精度的同时,也成功地保持了模型复杂性的合理平衡。Gait recognition is a biometric technology that identifies individuals through their unique walking patterns.It is suitable for unconstrained environments and has broad application prospects.Although current methods focus on using body part based representations,they often overlook the hierarchical dependencies among local motion patterns.In this paper,we propose a multi-level hierarchical fusion model for extracting gait features from coarse to fine.Our framework integrates two strategies:multi-level hierarchical feature extraction and non-uniform hierarchical feature extraction.This enables fine-grained extraction of local features while emphasizing the interrelationships among local features.Verified by numerous experiments on widely recognized datasets,the method we proposed has been proven to be effective.While improving the model accuracy,this method also successfully maintains a reasonable balance in model complexity.

关 键 词：局部特征 多级分层 不均匀 特征融合 步态识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]