基于轨迹模板匹配的动态手势识别方法

首先采用基于混合高斯模型与椭圆肤色模型进行手势分割,分割出手势区域,使用卡尔曼滤波器进行手势跟踪,获得手势中心点的位置.在此基础上,记录各帧中心点位置,得到运动轨迹,利用提出的轨迹模板匹配方法对动态手势进行识别.该方法利用基本的几何特征便可完成手势运动轨迹的设置与识别,无需特征选择或训练样本的搜集.最后,采用基于Zynq-7000的Zedboard平台对该算法进行实现,并采用HLS硬件加速工具进行算法加速.实验结果表明,该算法可实现较精确的手势识别,接受弹性的输入采样,识别正确率在95%以上,且通过硬件加速后,可在嵌入式平台中实时识别,具有较好的实时性.

Dynamic Hand Gesture Recognition Based on Track Template Matching

Firstly,the mixed Gaussian model and ellipse skin color model are used to segment the gesture region and then Kalman filter is used to track gesture,then the center position of the gesture is obtained.On the basis,the center position of each frame is recorded,and then the moving track is gotten.Next,the proposed track template matching method is used for dynamic gesture recognition.This method uses the basic geometric features to complete the setting and recognition of the trajectory of hand gestures,without the need of feature selection or collection of training samples.Finally,the algorithm is implemented on the Zedboard platform based on Zynq-7000 and the HLS hardware acceleration tool is used to accelerate the algorithm.The experiment results show that the proposed algorithm can achieve more accurate gesture recognition and accept flexible input sampling.The recognition rate is more than 95%,and it can be real-time recognized by the embedded platform after the hardware acceleration.So the algorithm has better real-time performance.