深度学习算法下多模态人体动作实时跟踪仿真  

Real-Time Tracking and Simulation of Multi-Modal Human Action under Deep Learning Algorithm

作　　者：刘丰平 杜远坤 侯惠芳[3] LIU Feng-ping;DU Yuan-kun;HOU Hui-fang(Cllege of Information Engineering,Zhengzhou University of Science and Technology,Zhengzhou Henan 450064,China;College of Big Data and Artificial Intelligence,Zhengzhou University of Science and Technology,Zhengzhou Henan 450064,China;College of Artificial Intelligence and Big Data,Henan University of Technology,Zhengzhou Henan 450001,China)

机构地区：[1]郑州科技学院信息工程学院,河南郑州450064 [2]郑州科技学院大数据与人工智能学院,河南郑州450064 [3]河南工业大学人工智能与大数据学院,河南郑州450001

出　　处：《计算机仿真》2024年第9期157-161,共5页Computer Simulation

基　　金：21年度河南省科技厅科技攻关项目(212102210138);22年度河南省科技厅科技攻关项目(222102210280)。

摘　　要：人体的运动姿态多样,光影和噪声等因素对人体运动特征的提取和分析具有很大影响。且人体动作跟踪过程若目标存在遮挡,易导致目标丢失问题,严重影响人体动作跟踪的精准度。为解决上述问题,引入深度学习方法中的自编码器,提出新的多模态人体动作实时跟踪方法。通过对人体动作图像采集处理,完成人体动作特征的粗提取;利用自编码器对粗特征增强,实现人体动作特征的细提取,建立常规人体动作特征模板;利用核相关滤波(kernel correlation filter,KCF)获取响应函数峰值,建立目标丢失判据,结合特征匹配算法,计算跟踪目标与区域化结构特征之间相似度,确定最佳跟踪目标,实现人体动作的实时跟踪。实验结果表明,所提方法的动作跟踪精准度高,时延短,表明所提方法的跟踪效果好。The diversity of human motion poses and the impact of factors such as light,shadow and noise on the extraction and analysis of human motion characteristics are significant.If there is an obstruction in the tracking process,it is easy to cause target loss,seriously affecting the accuracy of human motion tracking.In order to address these issues,an autoencoder in the deep learning method was introduced,and a real-time tracking method for multi-modal human motion was proposed.By processing human motion images,we extracted rough features of human motion.Then,we used the autoencoder to enhance these features,thus achieving the fine extraction of human motion features.Moreover,we constructed a template including conventional human motion features.Furthermore,we used kernel correlation filters(KCF) to obtain the response function peak and establish a target loss criterion.In combination with the feature matching algorithm,we calculated the similarity between the target and regional structural features,thus determining the best tracking target.Finally,we achieved the real-time human motion tracking.Experimental results show that the proposed method has high precision and short latency,and the tracking effect is good.

关 键 词：深度学习算法 多模态人体动作 实时跟踪方法 人体动作特征 跟踪算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]