【作者】 李海波；

【导师】 许洪光；

【作者基本信息】 哈尔滨工业大学 ， 电子与通信工程（专业学位）， 2016， 硕士

【摘要】 乒乓球作为我们的“国球”在人们的公共健身计划中广泛流行,但人们通常打乒乓球,需要其他人计分,人手不够时,需要自己分心去计分。本课题研究的目的就是利用计算机对乒乓球进行检测与轨迹跟踪,设计出一种能够实现比赛过程中自动计分的方法,使得计分更加便捷,以便解决传统人工收集工作需要时间和麻烦的问题,可以为人们后续的体育活动提供方便性与趣味性。本课题在阅读大量参考文献的基础上,通过对视频图像检测技术的分析研究,运用图像像素坐标系建立二维图像模型,设计并实现了完整的乒乓球自动计分的方法。本课题采用了手机摄像配合广角镜头的方法来获取有效的视频数据样本,对采集到的视频进行了格式转换及解帧处理。本课题使用Visual Studio和Open CV在视频解帧后对图像执行了一系列处理操作。将彩色图像进行了灰度化处理,在图像增强部分使用了中值滤波方法。在阈值分割处理中,本课题使用了大津法来决定在图像二值化过程中用到的阈值,然后对图像进行了膨胀与腐蚀等形态学操作。检测乒乓球时比较分析了三种方法,根据实际效果最终选用了帧间差分法与基于目标形状分析相结合的方法来实现对乒乓球的识别与定位。同时,利用Canny算子对乒乓球台进行了边缘检测,利用邻域搜索法实现了对乒乓球台轮廓坐标的提取。接着,通过设定有效区域,提取有效轮廓,再根据质心坐标公式求出了乒乓球在每帧图片中的质心位置坐标。最后,根据对乒乓球的质心位置坐标以及回合内的飞行轨迹分析对乒乓球的起始位置、飞行方向、出界、碰案以及下网等情况进行了相关判定,决策出相应的比分,将比分结果及原因进行了展示,达到了自动计分的效果。本课题针对检测乒乓球的算法及自动计分方法的准确率与误差存在的原因进行了分析,同时提出了改善准确率,减小误差的方法并做出了进一步的展望。更多还原

【Abstract】 Table tennis,the "national ball" in people’s public fitness program is widely popular in China.However,people usually play table tennis and need other people to score.When people are not enough,they need to be distracted to score.The purpose of this dissertation is to use the computer to detect and track the table tennis to design a method that can automatically score in the process of competition to make scoring more accurate and display the score in real-time to solve the problem of traditional manual collecting work.It not only costs time but also makes trouble.It provides convenience and fun for people in follow-up sport activities.Based on reading a lot of references,the analysis of video image detection technology,using the image pixel coordinate system to establish two dimensional image model,design and implement a complete table tennis automatic scoring algorithm.In this paper,we use the mobilephone camera with the wide-angle lens to obtain a valid video data samples.The format conversion and frame processing of the collected video are carried out.In this paper,we use Visual Studio and Open CV to process a series of image preprocessing.The color image is grayscale processed,and the median filtering method is used in the image enhancement part.In the threshold segmentation process,we use the Otsu method to determine the threshold of the image binarization,and then make the morphological operations such as erosion and corrosion to the image.In this paper,we compare and analysis three methods to find the ball.Finally,the method of interframe difference and target shape matching are used to realize the recognition and localization of table tennis according to the actual effect.At the same time,using Canny operator on the ping-pong table for edge detection,the use of neighborhood search method to achieve the ping-pong table contour coordinates extraction.Then,set up effective area and extract the effective outline,calculate the centroid coordinates of table tennis in every frame image according to centroid coordinates formula.Finally,according to the coordinates position of the center-of-mass of the ping-pong ball and the flight trajectory analysis in each turn,the relevant decisions about the starting position of table tennis,flight direction,out of bounds,the case of the hit and the ball under the net are determined.The results and the reasons for the display,to achieve the effect of automaticlly scoring.For the accuracy of algorithm for detecting table tennis and automatic scoring method,we analyzed the reasons of the existence of error,at the same time put forward for improving the accuracy and reduc ing the error.We make the further outlook.更多还原