改进的马氏距离动态时间规整手势认证方法

针对现有动态手势认证方法普遍采用的动态时间规整(DTW)算法在计算欧氏距离(ED)时各维特征向量被同等对待且各维特征间的相关性被忽视等问题,提出一种改进的马氏距离动态时间规整手势认证方法.通过手机内置三轴加速度传感器实时获取动态手势信号,经数据预处理后由改进动态时间规整算法进行加速度信号相似性度量,在计算过程中根据协方差矩阵特点进行时间复杂度优化,最后根据模板匹配法得出认证结论.实验结果表明,经改进后等错率(EER)由3.02%降至1.39%,经优化后认证响应时间平均降低87.84%.该方法进一步提高了动态手势认证精确度,同时实时性良好.

Dynamic time warping gesture authentication algorithm based on improved Mahalanobis distance

Concerning the problem that Euclidean Distance (ED) treats each feature equally and cannot take account of the correlations between features in Dynamic Time Warping (DTW) method which is commonly used in dynamic gesture authentication, a DTW gesture authentication algorithm based on improved Mahalanobis Distance (MD) was put forward. It utilized embedded three-axial accelerometer to capture the real-time dynamic hand signals, acceleration signal similarity was measured by the improved DTW method after data preprocessing, and according to the characteristics of the covariance matrix, the time complexity was optimized in the process of calculation, finally the result of authentication was got by template matching method. The experimental results show that the Equal Error Rate (EER) decreases from 3.02% to 1.39% and the average authentication response time reduces by 87.84% after optimization. The proposed method improves the accuracy of the dynamic hand gesture authentication and has a good real-time performance at the same time.