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%.
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)
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)。
