随着计算机技术的飞速发展，人们对人机交互的需求不断增加，利用计算机来理解人类行为，使人机交互更加简单化和人性化。手势识别作为一种直观、人性化的交互方式具有广泛的发展空间和市场需求，引起了许多研发人员的关注，已成为人机交互领域的研究热点。 本文以Kinect传感器为基础，主要研究了动态手势识别技术，设计并实现了一个可以完成动态手势识别和手势控制视频播放器两大功能的完整系统。本论文的研究主要从以下方面展开：首先，研究Kinect传感器获取深度数据和骨骼数据的原理，同时分析动态手势识别的具体方案，对各个模块用到的技术做了详细的研究。其次，分析手势检测方法并利用基于速度阈值的方法进行手势检测；在此基础上跟踪并记录左手、右手、左肘、右肘、肩部中心、左肩和右肩这七个关节点的骨骼三维坐标数据；同时，采用人体关节点相对距离系数作为动态手势识别算法的主要手势特征，构造出每帧4维的手势特征向量；分析常用的动态手势识别算法，重点研究了动态时间规整(DTW)算法,并作出搜索路径限制和距离加权两点改进，通过在限制寻优路径的同时对特征向量中的每个元素赋予不同权值，从而优化两个序列的匹配过程，在获取最优规整距离的基础上利用KNN算法得出识别结果。最后，搭建了一个完整的动态手势识别系统，定义了七种常用的交互手势，实现了视频播放器的实时控制功能。 实验结果表明：本文改进的DTW算法对预定义的七种手势平均识别率达96.3%，且在不同的背景和光照条件下有较强的鲁棒性。可以实现在实时条件下利用动态手势控制视频播放器的功能，完成简单的人机交互。

With the rapid development of computer technology, the demand for human-computer interaction is increasing. Using computers to understand human behavior, that makes human-computer interaction more simple and humanized. As an intuitive and humanized interaction method, gesture recognition has wide development space and market demand, which has attracted the attention of many researchers, and has become a hot research topic in the field of human-computer interaction. Based on Kinect sensor, this paper mainly studies the technology of dynamic gesture recognition, designs and implements a complete system which can realize the functions of dynamic gesture recognition and using gestures to control video player. The research of this thesis is mainly from the following aspects: Firstly, the principle of Kinect sensor in depth data and bone data acquisition is studied. At the same time, the specific scheme of dynamic gesture recognition is analyzed, and the technology used in each module is studied in detail. Secondly, the method of gesture detection is analyzed and the method based on velocity threshold is used to detect gesture. On this basis, we track and record the three dimensional coordinates of the seven joints of left hand, right hand, left elbow, right elbow, shoulder center, left shoulder and right shoulder.Taking the relative distance coefficient of human body joint as the main gesture features of the dynamic gesture recognition algorithm, the 4 dimensional gesture feature vector of each frame is constructed.we analyze the commonly used dynamic gesture recognition algorithm, focuse on the dynamic time warping (DTW) algorithm, and make search path restriction and distance weighted two point improvement, by limiting the optimization path and assigning different weights to each element in the feature vector,the matching process of two sequences is optimized. Based on obtaining the optimal regular distance, we use the KNN algorithm to get the recognition results. Finally, a complete dynamic gesture recognition system is built, and seven commonly used interactive gestures are defined to realize the real-time control function of the video player. The experimental results show that the improved DTW algorithm has an average recognition rate of 96.3 for the seven predefined gestures, and is robust under different background and illumination conditions. It can realize the function of controlling video player by dynamic gesture under real time condition, and complete simple human-computer interaction.