为提高基于加速度传感器的人体行为识别率，提出2种新的加速度数据特征。一种通过计算加速度矢量与重力方向夹角的小波能量来揭示加速度方向变化的本质，从时频分析的角度区分不同行为；另一种提取加速度数据重排后的关键点连线斜率，突出数据的差异和分布特点。将上述2种特征与常用的6种特征相结合，训练基于支持向量机的多类分类器，对7种日常行为进行识别。检测结果表明，独立检测法和留一交叉检测法对7种行为的平均识别率分别可达92.70%和95.08%。%Two novel features for acceleration data are applied to improve recognition accuracy of human activities. One feature uncovers the essential of acceleration direction by calculating the Wavelet Energy(WE) of angle between acceleration vector and gravity direction, and distinguishes different activities from time-frequency analysis. The other feature extracts from the slope of key points connection after acceleration data are rearranged, which highlights the difference and distribution of acceleration data. The two novel features can be combined with the six traditional widely used features to constitute feature sets, which allows to train the multi-class classifier based on Support Vector Machine(SVM), and to identify seven Activities of Daily Living(ADL). Two test results show that the average recognition accuracy of independent test method and leave one out cross test method can reach 92.70%and 95.08%respectively.
展开▼
