KinectV2传感器实时动态手势识别算法

KinectV2 Sensor Real-Time Dynamic Gesture Recognition Algorithm

为解决动态手势动作识别算法提取的特征泛化程度低,识别实时性差和识别率不高的问题,本文提出了一种基于KinectV2传感器的模板匹配手势识别方法。该方法利用Kinect Studio和Visual Gesture Builder建立手势动作数据库,提取人体关键骨骼间的夹角作为随时间变化的一维特征向量,借助动态规划思想,在动态时间规整(DTW)算法的基础上提出了加权DTW手势动作识别算法。通过对DTW算法的改进,使其规整路径更快的收敛,提高了手势动作识别的实时性和准确性,并在Unity 3D虚拟仿真实验平台中进行了实时性和有效性的验证实验,同时实现了简单的人机交互。

In order to solve the problems of low generalization of features extracted by the dynamic gesture action recognition algorithm, poor real-time recognition and low recognition rate, this paper proposes a template matching gesture recognition method based on KinectV2 sensor. This method uses Kinect Studio and Visual Gesture Builder to build a gesture action database, extracts the angle between key bones of the human body as a one-dimensional feature vector that changes with time, and proposes Weighted DTW gesture motion recognition algorithm based on the dynamic time warping (DTW) algorithm with the help of dynamic planning ideas. Through the improvement of the DTW algorithm, the regular path converges faster, the real-time and accuracy of gesture motion recognition is improved, and the real-time and validity verification experiments are carried out in the Unity 3D virtual simulation experiment platform for simple human-computer interaction.