基于视频图像处理的网球发球行为分析

发布时间：2018-01-24 17:14

本文关键词： 网球发球稀疏表示最佳击球点支持向量机粒子群算法网球发球模型　出处：《浙江理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：基于视频的人体运动分析是计算机视觉领域的一个重要研究方向,它是从视频序列中检测出运动物体,提取出人体关键部位,获得人体运动的有用信息,实现对人体动作、姿态等的进一步分析和识别。传统的网球发球训练模式已经根深蒂固的存在于现在的训练之中。对于运动员来说,要想熟练地掌握技术动作要领就必须根据教练员的指导意见进行长时间的,反复的练习。长时间的采用以经验为主的训练方法,凭借教练员的主观意识对运动员技术动作进行指导和监督,这种情况严重制约了网球水平的提高。针对此问题,本文作者在充分了解相关研究工作的基础上,以网球的平击发球为研究背景,主要研究内容如下:(1)为了提高网球平击发球的准确率和训练效率。本文首先对发球手臂的关节点进行颜色标记,通过高速摄像机采集网球发球视频,以每一帧图片中标记点坐标代替关节点坐标,来研究发球手臂的运动轨迹。在视频处理过程中,对一系列噪声图构造字典后,由稀疏表示思想重构出无干扰的发球图,采用混合高斯背景建模进行运动前景提取。获得运动前景后,通过颜色特征提取出标记点,并对其进行二值化操作,接着进行标记点轮廓查找,用最小圆包围轮廓,返回的圆心坐标作为关节点坐标。实验数据表明,击球点和身高的比值在[1.37,1.44]范围内,发球成功率达到97.24%。(2)为了对标记点轨迹进行进一步研究,本文将肩部标记点的发球轨迹作为研究对象,建立了基于改进支持向量机的网球发球模型。针对传统的支持向量机没有解决模型中惩罚参数c和核函数的参数值σ如何确定的问题,本文以分类精度作为适应度函数,采用粒子群算法对其进行改进。通过用训练数据对该模型进行训练之后,返回最优的惩罚参数c和核参数σ。实验表明,该模型可以准确可靠地对发球轨迹进行分类,其准确率为97.5%,接着对该模型与传统SVM进行了对比分析,PSO_SVM的分类误差要远小于SVM。
[Abstract]:Human motion analysis based on video is an important research direction in the field of computer vision. It is used to detect moving objects from video sequences, extract key parts of human body and obtain useful information of human motion. To achieve the further analysis and recognition of human body movements and posture. The traditional tennis serve training mode has been deeply rooted in the current training. For athletes. In order to master the key points of technical action skillfully, it is necessary to practice for a long time and repeatedly according to the guidance of the coaches, and to adopt the experience-oriented training method for a long time. Relying on the subjective consciousness of coaches to guide and supervise the athletes' technical movements, this situation seriously restricts the improvement of tennis level. In view of this problem, the author of this paper fully understand the relevant research work on the basis. In order to improve the accuracy and training efficiency, this paper first carries on the color marking to the key node of the serve arm in order to improve the accuracy and the training efficiency of the tennis flat stroke serve as the research background, the main research content is as follows: 1) in order to improve the accuracy and training efficiency of the tennis flat stroke serve. The video of tennis serve is collected by high speed camera, and the coordinate of mark point in each frame is replaced by the coordinate of the node to study the track of the serve arm. In the process of video processing. After constructing a dictionary of a series of noise images, the non-interference serve map is reconstructed from sparse representation idea, and the motion foreground is extracted by using mixed Gao Si background modeling. The marking points are extracted from the color features and binary operation is carried out. Then the marking points are searched and the contour is surrounded by the smallest circle. The returned circular center coordinates are taken as the node coordinates. The experimental data show that. The ratio of hitting point to height is. [In the range of 1.37 / 1.44, the success rate of serve reached 97.24.2) in order to further study the track of the mark point, this paper takes the track of the shoulder mark point as the research object. A tennis serve model based on improved support vector machine (SVM) is established. The traditional SVM does not solve the problem of how to determine the penalty parameter c and the parameter value 蟽 of kernel function. In this paper, classification accuracy is taken as fitness function, particle swarm optimization algorithm is used to improve it. After training the model with training data, the optimal penalty parameter c and kernel parameter 蟽 are returned. The model can classify the serve track accurately and reliably, and the accuracy is 97.5. then the model is compared with the traditional SVM. The classification error of PSO_SVM is much smaller than that of SVM.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：G845;TP391.41

【相似文献】