一种基于空间角度序列递归模型的动态手势识别算法

Dynamic Gesture Recognition Based on Space Angle Sequence Recurrence

针对传统动态手势识别方法准确性不高,鲁棒性不强的问题,研究动态手势轨迹运动的行为特征,从手势运动的轨迹点角度出发提出了一种基于递归图压缩的CK-1手势相似性检测算法,称为Tra DotRP(Gesture Recognition based on the Trajectory Dots Sequence Recurrence Plot).首先利用阈值分割方法获取kinect传感器采集到的手势运动轨迹点序列,计算轨迹点与初始点连线的方向角,然后对手势轨迹进行方向角的特征序列提取;为了克服手势轨迹序列不等长问题,构造基于时空域的手势轨迹序列递归图;再利用MPEG-1压缩算法计算手势递归图之间的CK-1距离,最终完成动态手势识别目标.实验结果表明,该方法可以有效地解决手势轨迹序列不等长问题,在120个动态手势的数据集上准确率高达97.48%,超过目前流行的手势识别算法.

To address the accuracy and robustness problem of traditional gesture recognition algorithms,by studying the behavior characteristics of dynamic hand gesture,a novel algorithm called Trajectory Dots Sequence Recurrence Plot（ Tra DotRP） is proposed based on compression distance of recurrence patterns from the perspective of space angle sequence. Firstly,the method of threshold segmentation is adopted to obtain the trajectory dots sequence of the hand using Kinect sensor,and then the orientation angle sequence characters are obtained through the track points and the initial points. In order to overcome the problem of unequal length of the trajectories sequence,recurrence plots in spatial domain is established; Finally,CK-1 distance is calculated using the MPEG-1 video compression algorithm for gesture recognition. The experimental results show that the proposed algorithm achieves high robustness in the case of the unequal length of hand trajectory sequence,with 97. 48% recognition accuracy on the data set of 120 dynamic gestures,which already overtaken the current popular hand gesture recognition algorithms.