针对现有的手势识别算法识别率低、鲁棒性弱的问题，提出一种基于Kinect骨架信息的交通警察手势识别方法。从Kinect深度图像中预测人体骨架节点的坐标位置，将节点的运动轨迹作为训练和测试的特征，结合距离加权动态时间规整算法和K-最近邻分类器进行识别。实验表明，在参数最优的情况下，该方法对八种交通警察手势的平均识别率达到98.5%，可应用于智能交通等领域。%To overcome the problems such as low recognition rate and weak robustness of current gesture recognition algorithms, this paper presents a novel traffic gesture recognition method based on Kinect skeleton data. It predicts skeleton joints’coordinates from depth image captured by Kinect sensor. Then it uses joints’trajectories as the features of training and testing. Distance weighting dynamic time warping algorithm and K-nearest neighbor algorithm are used to recognize the giving sample. The experimental results show that when argument is optimal, the average recognition rate is up to 98.5%tested on eight traffic cop gestures, so this method can be applied to intelligent traffic field.
展开▼
