基于线性加速度的多节点人体行为识别

针对当前基于加速度人体行为识别方法中存在的行为数据易受重力加速度影响以及空间信息欠缺等问题,提出一种基于线性加速度的多节点人体行为识别算法。通过分段双向去除重力加速度算法,去除传感器加速度中的重力加速度得到线性加速度;使用滑动均值滤波器滤除线性加速度与传感器加速度的颤抖运动,并对两种加速度中的冗余动作进行裁剪;分别从两种加速度中提取不同关节点数据间的动态时间规整算法(dynamic time warping, DTW)距离特征以及7种常规时域特征;利用支持向量机对人体行为进行分类。试验结果表明,该方法能有效提高人体行为识别的准确性。

Multi-node human behavior recognition based on linear acceleration

Focused on the issue that the behavior data in the current acceleration-based human behavior recognition method was affected by the gravitational acceleration and the lack of spatial information, a multi-node human behavior recognition algorithm based on linear acceleration was proposed. The linear acceleration was obtained by removing gravitational acceleration using segmented bi-directionally removal of gravitational acceleration algorithm. The tremor motion signal was filtered by a sliding averaging filter for linear acceleration and sensor acceleration, and the redundant actions in the two accelerations were cropped. The dynamic time warping (DTW) distance features between different joint points and seven conventional time domain features were extracted from two accelerations. The support vector machine was employed to recognize the human behavior. Experimental results showed that this method could effectively improve the accuracy of human behavior recognition.