基于Kinect的人体动作识别算法研究

针对现有的复杂背景下人体动作识别中存在识别准确率不高和实时性不强等问题，提出基于Kinect骨骼数据的改进动作识别算法。通过Kinect获取骨骼数据，提取出人体关节的特征向量，然后用模板匹配的方法对人体动作进行识别。通过搭建机器人体感控制系统验证了算法的可行性。在相同实验条件下测得算法的平均识别率为95.2%，平均识别时间为32.5ms。与其它动作识别算法比较，证明了算法的识别率较高、实时性较好。

Study on Motion Recognition Algorithm Based on Kinect

Under the complex background, with the problem of low accuracy and real-time?of the motion recognition algorithm, an improved algorithm was proposed based on Kinect skeleton data. Getting skeleton data through the Kinect, feature vector of human joints was extracted and then motions were recognized by using the method of template matching. The feasibility of the algorithm was verified by building the robot gesture?control system.Under the same experimental conditions, the average recognition rate of the algorithm is 95.2% and the average recognition time is 32.5ms. Compared with other motion recognition algorithm, higher recognition rate and better? real-time??property of the algorithm were proved.