基于多特征融合的活动识别算法  

Activity Recognition Algorithm Based on Multi-feature Fusion

作　　者：张耀威 李瑞祥[1] 戴健威 施伟斌[1] ZHANG Yao-wei;LI Rui-xiang;DAI Jian-wei;SHI Wei-bin(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《软件导刊》2022年第3期72-77,共6页Software Guide

摘　　要：基于可穿戴传感器的人体活动识别技术在许多领域应用广泛,从人体活动信号中提取丰富的特征是提高活动识别准确率的关键技术之一。为此,提出基于傅里叶描述子(FDs)、局部二值特征(LBP)和小波能量谱(WES)的融合特征提取人体活动的详细信息。为提高识别系统的可靠性,去除对识别精度没有影响的冗余特征,引入过滤式选择算法Relief-F进行特征选择,筛选对不同活动具有较高区分度的特征,然后利用随机森林分类器对多种不同活动进行精确识别。基于Python3.6平台,在公开的WISDM和ADL数据集上验证该算法的有效性。实验结果表明,多特征融合算法对WISDM和ADL数据集分别取得了94.5%和95.3%的识别准确率,识别效果明显优于单一特征算法,具有很强的鲁棒性。Human activity recognition technology based on wearable sensors has been widely used in many fields.Extracting rich features from human activity signals is one of the keys to improve the accuracy of activity recognition.In this paper,a fusion feature based on Fourier descriptor(FDs),local binary pattern(LBP)and wavelet energy spectrum(WES)is proposed to extract detailed information of human activities.In order to improve the reliability of the recognition system and remove redundant features that have no impact on the recognition accuracy,Relief,a filtering selection algorithm,is introduced in this paper to select features,and features with high discrimination for different activities are selected.Finally,random forest(RF)classifier is used to accurately identify a variety of different activities.Based on Python3.6 platform,the effectiveness of the algorithm is verified on the public WISDM and ADL dataset.Experimental results show that the multi-feature fusion algorithm recognition accuracy of WISDM and ADL is 94.5%and 95.3%,respectively,which has better recognition performance than the single feature algorithm,and has strong robustness.

关 键 词：傅里叶描述子 局部二值特征 特征选择 人体活动识别 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]