[1]陈 鑫,叶 宁,徐 康,等.基于 EfficientNet 模型的毫米波雷达人体行为识别[J].计算机技术与发展,2022,32(09):134-141.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 021]
　CHEN Xin,YE Ning,XU Kang,et al.FMCW Radar Human Action Recognition System Based on EfficientNet Model[J].,2022,32(09):134-141.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 021]
点击复制

基于 EfficientNet 模型的毫米波雷达人体行为识别()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
32
期数:
2022年09期
页码:
134-141
栏目:
人工智能
出版日期:
2022-09-10

文章信息/Info

Title:
FMCW Radar Human Action Recognition System Based on EfficientNet Model
文章编号:
1673-629X(2022)09-0134-08
作者:
陈 鑫1; 2 ; 叶 宁1; 2 ; 徐 康1; 2 ; 王 甦1; 2 ; 王汝传1; 2
1. 南京邮电大学 计算机学院、软件学院、网络空间安全学院,江苏 南京 210023
2. 江苏省无线传感网高技术研究重点实验室,江苏 南京 210023
Author(s):
CHEN Xin1; 2 ; YE Ning1; 2 ; XU Kang1; 2 ; WANG Su1; 2 ; WANG Ru-chuan1; 2
1. School of Computer Science,School of Software,School of Cyberspace Security,Nanjing University of Posts and Telecommunications,Nanjing 210023,China
2. Jiangsu High Technology Research Key Laboratory for Wireless Sensor Networks,Nanjing 210023,China
关键词:
调频连续波雷达; 行为识别; 距离多普勒; EfficientNet; 深度学习; 机器学习; 迁移学习
Keywords:
FMCW radar; action recognition; range Doppler; EfficientNet; deep learning; machine learning; transfer learning
分类号:
TP391. 4
DOI:
10. 3969 / j. issn. 1673-629X. 2022. 09. 021
摘要:
针对基于传统光学摄像头实现人体行为识别系统所带来的隐私暴露,易受光照、遮挡的影响等问题,设计基于EfficientNet 模型的 FMCW 雷达人体行为识别系统。 首先, 对 FMCW 雷达实测数据采用距离多普勒算法构建每一帧距离-速度图像; 接着,采用逐帧积累的方法延长观测时间窗口,构建覆盖整个动作过程的距离-速度轨迹;最后采用改进预训练的EfficientNet 模型对不同人体行为进行识别。 实验结果表明,在 5 秒观测时间窗口内,改进预训练的 EfficientNet-B4 模型对已知个体和未知个体 9 种行为识别准确率达到 99. 3% 与 98. 2% ,均高于传统机器学习方法及经典深度学习方法,进一步缩短观测时间窗口至 2. 5 秒,改进预训练的 EfficientNet-B4 模型对已知个体和未知个体的 9 种行为识别准确率仍能达到96. 7% 与 95. 4% 。 除此之外, 在 5 秒观测时间窗口内,所提方法对已知个体和未知个体的 9 种行为识别准确率比常见利用时间-速度提取行为参数的方法分别提高了 3. 5% 与 4. 9% ,缩短观测时间窗口至 2. 5 秒,所提方法准确率提高了 4. 2% 与4. 8% ,可见所提方法可以有效地提升 FMCW 雷达人体行为识别的准确率,且模型的泛化能力较强。
Abstract:
Since using traditional optical camera to realize human action recognition system will bring some problems,such as privacy exposure,easy to be affected by light and occlusion, a FMCW radar human action recognition system based on Efficient Net model isdesigned. Firstly,range Doppler algorithm is used to construct range-velocity images of each frame from the measured data of FMCWradar. Then the range velocity trajectory covering the whole movement process is constructed by using the method of frame-by-frame accumulation to extend the observation time window. Finally,the improved pre-training Efficient Net model is used to recognize different human actions. The experimental results show that within the 5 - second observation time window,the accuracy of the improved pre -training EfficientNet-B4 model for identifying nine actions of known and unknown individuals is 99. 3% and 98. 2% ,which are both higher than that of traditional machine learning methods and classical deep learning methods. When the observation time window was further shortened to 2. 5 seconds,the recognition accuracy of the improved pre trained efficientnet-b4 model for 9 behaviors of known and unknown individuals could still reach 96. 7% and 95. 4% . In addition,within the 5-second observation time window,the accuracy of the proposed method for identifying 9 actions of known and unknown individuals is 3. 5% and 4. 9% higher than the common methods of extracting behavior parameters by time - velocity. When the observation time window is shortened to 2. 5 seconds, the accuracy of theproposed method is improved by 4. 2% and 4. 8% . It can be seen that the proposed method can effectively improve the accuracy of FMCW radar human action recognition,and the model has strong generalization ability.

相似文献/References:

[1]王博,李燕.视频序列中的时空兴趣点检测及其自适应分析[J].计算机技术与发展,2014,24(04):49.
　WANG Bo,LI Yan.Space-time Interest Points Detection in Video Sequence and Its Adaptive Analysis[J].,2014,24(09):49.
[2]高蕾[],曹建忠[]. 基于可穿戴传感器的行为识别随机逼近模型[J].计算机技术与发展,2014,24(12):83.
　GAO Lei[],CAO Jian-zhong[]. Activity Recognition Using Stochastic Approximation Model Based on Wearable Sensor[J].,2014,24(09):83.
[3]石爱辉,程勇,曹雪虹.结合码本优化和特征融合的人体行为识别方法[J].计算机技术与发展,2018,28(02):107.[doi:10．3969/j．issn．1673－629X．2018．02．024]
　SHI Ai-hui,CHENG Yong,CAO Xue-hong.A Human Action Ｒecognition Method Combined with Codebook Optimization and Feature Fusion[J].,2018,28(09):107.[doi:10．3969/j．issn．1673－629X．2018．02．024]
[4]黄菲菲[],曹江涛[],姬晓飞[]. 基于多通道信息融合的双人交互动作识别算法[J].计算机技术与发展,2016,26(03):58.
　HUANG Fei-fei[],CAO Jiang-tao[],JI Xiao-fei[]. Two-human Interaction Recognition Algorithm Based on Multi-channels Information Fusion[J].,2016,26(09):58.
[5]毛驾燕,沈苏彬.一种面向智能家居老人看护系统的实现方案[J].计算机技术与发展,2018,28(08):164.[doi:10.3969/ j. issn.1673-629X.2018.08.035]
　MAO Jia-yan,SHEN Su-bin.A Realization Scheme for Elderly Care System in Smart Home[J].,2018,28(09):164.[doi:10.3969/ j. issn.1673-629X.2018.08.035]
[6]韩欣欣,叶奇玲.基于 SIFT 和 HOG 特征融合的人体行为识别方法[J].计算机技术与发展,2019,29(06):71.[doi:10. 3969 / j. issn. 1673-629X. 2019. 06. 015]
　HAN Xin-xin,YE Qi-ling.Human Action Recognition Based on Feature Fusion of SIFT and HOG[J].,2019,29(09):71.[doi:10. 3969 / j. issn. 1673-629X. 2019. 06. 015]
[7]闫 坤,沈苏彬.一种基于智能家居的用户行为预测方法[J].计算机技术与发展,2020,30(01):19.[doi:10. 3969 / j. issn. 1673-629X. 2020. 01. 004]
　YAN Kun,SHEN Su-bin.A User Behavior Prediction Method Based on Smart Home[J].,2020,30(09):19.[doi:10. 3969 / j. issn. 1673-629X. 2020. 01. 004]
[8]杜潘飞,王志辉,李雄伟,等.异常行为检测数据集快速构建方法[J].计算机技术与发展,2021,31(09):155.[doi:10. 3969 / j. issn. 1673-629X. 2021. 09. 026]
　DU Pan-fei,WANG Zhi-hui,LI Xiong-wei,et al.Fast Construction Strategy of Abnormal Action Detection Dataset[J].,2021,31(09):155.[doi:10. 3969 / j. issn. 1673-629X. 2021. 09. 026]
[9]姚 捃,郭志林.一种端到端的考场多目标行为识别算法[J].计算机技术与发展,2022,32(09):174.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 027]
　YAO Jun,GUO Zhi-lin.An End-to-end Multi-objective Behavior Recognition Algorithm for Examination Room[J].,2022,32(09):174.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 027]
[10]李嘎,加云岗,王志晓,等.基于YOLO-CARAFE的人员异常行为识别方法[J].计算机技术与发展,2024,34(06):185.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0093]
　LI Ga,JIA Yun-gang,WANG Zhi-xiao,et al.Human Abnormal Behavior Recognition Method Based on YOLO-CARAFE[J].,2024,34(09):185.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0093]

更新日期/Last Update: 2022-09-10