Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (21): 205-212.DOI: 10.3778/j.issn.1002-8331.2105-0129

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Joint Location and Behavior Recognition in Deep Residual Shrinkage Network

ZHANG Li, CHANG Jun, WU Hao, HUANG Bin, LIU Huan   

  1. College of Information, Yunnan University, Kunming 650500, China
  • Online:2022-11-01 Published:2022-11-01



  1. 云南大学 信息学院,昆明 650500

Abstract: Body sensing methods based on WiFi channel state information(CSI) have been applied in many Internet of things scenarios, but most existing CSI body sensing systems only perform one of the tasks of positioning or behavior recognition. The development of the Internet of things requires them to be able to identify simultaneously. To solve this problem, a joint localization and behavior recognition method based on deep residual shrinkage network is proposed. CSI data of two scenes(dark room and conference room) are obtained through common commercial WiFi devices. Combined with the pre-processed data input and the learning model of deep residual shrinkage network, joint task identification is performed for 12 positions and 6 daily behaviors(standing up, sitting down, jumping, squatting, falling down, and picking up). The experimental results show that the average recognition rate for indoor positioning in dark room, meeting room and corridor is 97.29%, and the average recognition rate for behavior recognition is 90.02%. It can realize high precision joint identification of location and behavior.

Key words: channel state information, behavior recognition, indoor positioning, joint identification, deep residual shrinkage network, soft threshold

摘要: 基于WiFi信道状态信息(channel state information,CSI)的人体感知方法在许多物联网场景得到了应用,但现有大部分基于CSI人体感知的系统仅进行定位或行为识别其中一项工作,而物联网的发展对两者能同时识别提出了新的要求。针对这一问题,提出一种基于深度残差收缩网络的定位与行为联合识别方法。通过普通商用WiFi设备获取两种场景(暗室、会议室和走廊)的CSI数据,将预处理后的数据输入结合了深度残差收缩网络的学习模型,进行12个位置与和6种日常行为(站起、坐下、跳跃、深蹲、跌倒、捡起)的联合任务识别。实验结果显示,针对在暗室、会议室和走廊三种场景下的室内定位的平均识别率达到97.29%,行为识别的平均识别率达到90.02%。能够实现定位与行为的高精度联合识别。

关键词: 信道状态信息, 行为识别, 室内定位, 联合识别, 深度残差收缩网络, 软阈值化