结合形状序列与LTN的步态身份识别

Gait identification combined with shape sequence and LTN

为提高现有基于智能手机加速度传感器的步态身份识别方法的性能,提出一种基于典型步态周期提取的身份识别方法。针对现有方法中存在的周期检测错误和周期间的相位偏差问题,采用形状上下文和线性时间归一化(linear time normalization,LTN)相结合的方法对步态周期做序列校准匹配,从中提取典型步态周期来表征整个步态,为检测该方法的性能,采集40个志愿者的步态数据,利用1NN分类算法完成身份识别。实验结果表明,SC-LTN算法的平均正确识别率达96%,验证该方法提取的典型周期能有效用于身份识别。

To improve the performance of the existing gait identification method based on the intelligent mobile accelerometer,an identification method based on typical gait cycle was proposed.Aiming at the problem of periodic detection errors and the deviation of the cycle phase in the existing methods,the gait period was matched using the method of shape context and linear time normalization(LTN),and the typical gait period was extracted to characterize the whole gait.To detect the performance of the proposed method,the gait data of 40 volunteers were collected,and the identification was accomplished using 1NN classification algorithm.Experimental results show that the average correct recognition rate of SC-LTN algorithm is 96%,which verifies that the typical period of the proposed method is effective for identification.