摘要详情

ID / 提交时间

239 / 2020-01-03 15:20:00

标题

A Real-time Human Activity Recognition Method for TWR

关键字

Human Activity Recognition; through-the-wall radar; real-time; long short time memory network

主题及专题

IEEE SAM 2020: Full Regular Papers > Artificial intelligence in array processing

状态

全文被拒

作者

Can Cheng / University of Electronic Science and Tech, China

Fei Ling / University of Electronic Science and Tech, China

Shisheng Guo / University of Electronic Science and Technology of China, China

Guolong Cui / University of Electronic Science and Technology of China (UESTC), China

Lingjiang Kong / University of Electronic Science and Technology of China (UESTC), China

Chao Jia / University of Electronic Science and Tech, China

Xiaobo Yang / University of Electronic Science and Technology of China, China

摘要

Human activity recognition (HAR) has long been a question of great interest in anti-terrorism and other applications. In recent years, radar has been one of
the most widely used sensing modality of detecting human motion and have been extensively used for HAR. This paper researches on blocked human activity description method and deep learning recognition algorithm. In order to obtain the real-time nature of recognition under the condition of through-wall probing scenario, we proposes a range profile sequence driven FC-SLSTM-FC end-to-end model, which employ random crop training method. Based on the architecture above, we process the actual radar data, finally achieve an average accuracy of 97.6%
and real-time output with delay of only milliseconds.