改进GaitSet模型的煤矿井下人员步态识别方法  

Gait recognition algorithm of underground coal mine personnel based on improved GaitSet model

作　　者：汝洪芳 赵晖 王国新 Ru Hongfang;Zhao Hui;Wang Guoxin(School of Electrical&Control Engineering,Heilongjiang University of Science&Technology,Harbin 150022,China)

机构地区：[1]黑龙江科技大学电气与控制工程学院,哈尔滨150022

出　　处：《黑龙江科技大学学报》2025年第2期301-306,共6页Journal of Heilongjiang University of Science And Technology

基　　金：黑龙江省重点研发计划指导类项目(GZ20220122);黑龙江省省属高等学校基本科研业务费项目(2021-KYYWF-1480)。

摘　　要：针对步态识别模型准确率低与步态特征提取不充分的问题,提出一种改进GaitSet模型的煤矿井下人员步态识别方法。在GaitSet模型的基础上,引入多尺度卷积神经网络进行特征提取,采用多级池化模块,以保留主要的步态特征,提升模型的泛化能力,在CASIA-B数据集和自建煤矿井下人员步态数据集上进行验证。结果表明,排除相同视角后,三种状态下的平均识别准确率分别提升了0.53%、2.06%和1.35%,在自建煤矿井下人员数据集上,平均识别准确率提升了3.63%。This paper is aimed at addressing the low accuracy and insufficient feature extraction in gait recognition models,and proposes a gait recognition method of underground coal mine personnel based on an improved GaitSet model.The study consits of introducing a multi-scale convolutional neural network for the feature extraction on the basis of the GaitSet model,adopting a multi-level pooling module for the retention of the the main gait features,enhancing the generalization ability of the model,and verifying the CASIA-B dataset and the self-built underground coal mine personnel gait dataset.The results show that after excluding the same perspective,the average recognition accuracy under the three states increases by 0.53%,2.06%,and 1.35%respectively.And in the self-built underground coal mine personnel dataset,the average recognition accuracy increases by 3.63%.

关 键 词：煤矿 步态识别 GaitSet 多尺度卷积 多级池化 

分 类 号：TD76[矿业工程—矿井通风与安全] TP181[自动化与计算机技术—控制理论与控制工程]