机器人乒乓球的快速识别与跟踪研究

发布时间：2018-03-09 23:19

本文选题：乒乓球机器人　切入点：视觉系统　出处：《山东大学》2017年硕士论文　论文类型：学位论文

【摘要】：飞行物的识别与跟踪在国防、航空航天、工业与服务业等领域均具有重大的研究价值与应用价值,其中涉及到的实时视觉感知技术也是机器人的关键技术。本文以乒乓球机器人为研究平台,以飞速运动的乒乓球为研究对象,重点开展乒乓球机器人高速实时视觉技术的研究。在总结前人系统与技术的基础上,本文的主要研究内容包括:1、从摄像机的选型、立体视觉方案的选择与高速采集方案的设计三个方面详细介绍了高速彩色双目立体视觉系统搭建的过程。从输出信号、输出接口与相机芯片三个不同的方面进行对比,发现由CMOS感光芯片制作的USB3.0工业相机更具有优势。然后从性价比方面考量了多家厂商的不同型号的工业相机,最终选择Balser公司的aca1920-155uc型号的工业相机。通过对比单目、双目与多目视觉系统,双目视觉系统在此场景中是最优的选择。在分析通用采集方案的基础上,对原有采集方案进行改进,得到一种可以满足实时性与同步性要求的高速采集方案。最终搭建的系统视野范围覆盖整个乒乓球桌,可以实现对高速飞行的乒乓球运动轨迹的高速采集。2、在双目视觉系统标定方面,根据张正友平面标定原理,结合Matlab标定工具箱,分别对左右摄像机的焦距、主点构成的内参数和旋转矩阵、平移矩阵构成的外参数进行标定。此方法降低了标定误差,提高了检测精度,完全满足双目立体视觉系统的需求。3、详细介绍了平行双目视觉模型、三角测距与立体校正Bouguet算法,在此基础上描述了目标深度信息的计算过程。4、根据乒乓球的固有特点与乒乓球运动背景不变的特点,通过对比基于颜色的分割算法、帧差法与背景减除法,发现背景减除法作为目标区域初次分割算法表现最佳。为了提高检测精度,对初次分割的目标区域进行开闭运算和Canny算子边缘检测操作。根据乒乓球固有的特征,采用随机霍夫圆变换对目标物体进行识别,得到目标中心位置。为了提高跟踪的实时性,提出了基于上一帧目标局部区域搜索算法。最终实现了高速飞行的乒乓球三维位置空间的实时检测与跟踪。最后,实验验证了本文所提出的目标识别与跟踪算法的正确性与快速性,并进行了总结与展望。
[Abstract]:The identification and tracking of flying objects have great research value and application value in the fields of national defense, aerospace, industry and service industry, etc. The real-time visual perception technology involved is also the key technology of robot. This paper takes the table tennis robot as the research platform and the fast moving table tennis as the research object. On the basis of summarizing the previous systems and technologies, the main research contents of this paper include: 1, from the selection of camera, This paper introduces in detail the process of setting up a high speed color binocular stereo vision system from three aspects: the selection of stereo vision scheme and the design of high speed acquisition scheme. The output signal, output interface and camera chip are compared in three different aspects, such as the output signal, the output interface and the camera chip. It is found that the USB3.0 industrial camera made by CMOS photosensitive chip has more advantages. Then, from the aspect of performance and price ratio, we consider different models of industrial cameras from many manufacturers, and finally choose the industrial camera of aca1920-155uc model of Balser Company. Binocular vision system and multi-eye vision system, binocular vision system is the best choice in this scene. Based on the analysis of the general collection scheme, the original collection scheme is improved. A high speed acquisition scheme which can meet the requirements of real-time and synchronism is obtained. Finally, the whole table tennis table is covered by the system visual field. In the aspect of calibration of binocular vision system, according to the calibration principle of Zhang Zhengyou plane, combined with Matlab calibration toolbox, the focal length of left and right cameras can be achieved. The inner parameters and rotation matrix of the main points and the external parameters of the translation matrix are calibrated. This method reduces the calibration error and improves the detection accuracy. Fully meet the requirements of binocular stereo vision system. The parallel binocular vision model, triangulation and stereo correction Bouguet algorithm are introduced in detail. On this basis, the calculation process of target depth information is described. According to the inherent characteristics of table tennis and the invariant characteristics of table tennis movement background, the color-based segmentation algorithm, frame difference method and background subtraction method are compared. It is found that the background subtraction algorithm performs best as the initial segmentation algorithm of target area. In order to improve the detection accuracy, the opening and closing operations and the Canny operator edge detection operation are carried out for the initial segmentation of the target region. According to the inherent characteristics of table tennis, In order to improve the real-time tracking, the target center position is obtained by using random Hough circle transform to identify the target object. A local region search algorithm based on the previous frame is proposed. Finally, the real-time detection and tracking of the high speed flight table tennis 3D position space is realized. Finally, Experiments verify the correctness and rapidity of the proposed target recognition and tracking algorithm.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【相似文献】