【作者】 王飞；

【导师】 乌日图；

【作者基本信息】 内蒙古工业大学 ， 机械电子工程， 2007， 硕士

【摘要】 双目立体视觉是机器人视觉技术的一个重要分支,即不同位置的两台摄像机或一台摄像机经过移动或旋转拍摄同一场景,通过计算空间点在两幅图像中的视差,获取该点的空间位置信息。基于双目视觉平台的研究主要包括摄像机的标定、图像处理、立体匹配、三维重建以及运动分析等。通过对双目立体视觉技术的进一步研究,为解决机器人的视觉问题如智能机器人的导航问题、工业机器人的视觉检测以及手臂抓取等问题奠定了技术基础。摄像机标定与图像采集。运用大恒公司的DH-CG300图像采集卡循环采集两帧图像到内存,该部分内容涉及到图像卡的初始化及如何打开图像采集卡、开始采集、停止采集、结束采集、摄像机的标定几部分内容。对于摄像机的标定,课题中引入了一种摄像机模型,该模型有11个参数。利用6个已知其空间坐标和图像坐标的点,将6个控制点代入到摄像机的成像模型中,构成一个齐次方程组,然后求解出各个参数。至此实现了摄像机的标定和图像的采集。对采集到内存中的图像进行处理。首先,将机械手要抓取的目标物做成一个模板,模板的制作要尽可能的小,能够包含目标物的主要特征即可。然后,利用模板匹配的方法确定实时图像当中目标物所在的位置,并记下目标物中心点的图像坐标,图像坐标系的原点设定为图像的左下角。最后,将匹配出的目标物所在的区域显示到控制界面上。该部分内容实现了目标物中心点图像坐标的确定。实现机械手的运动控制。目标物的空间坐标确定以后,其相对于机械手当前的位置也就确定,由此可以得到机械手运动到目标物所在位置所需的进给量。控制机械手运动到目标物所在位置,抓取目标物,并放到指定的位置,放置的位置可以由用户自己设定。本设计实现了用图像处理的方法来控制机械手抓取目标物,控制精度较高,适用性较广。硬件实现简单,成本较低。更多还原

【Abstract】 Two-camera stereo visual system is an important branch of robotic visual technologies. It means that two cameras on different locations or one camera from dissimilar view angles via translating or rotating shoot the same scene to get the parallax according to which the space position information to the object center point is calculated. The study of two-camera visual platforms includes cameras calibration, image processing, stereo matching, three-dimensional reconstruction and movement analysis. Through the further research of two-camera visual technologies, it establishes technical foundations of robotic vision problems such as smart mobile robot navigation and industrial robots visual detection and manipulator grabbing.We use the image collection card DH-CG300 which is made by Daheng company circulation collect two frames of images arriving at memory, this part of content includes initialization of the image card,opening the image collection, starting collection, stopping collection, ending collection and camera calibration. Toward the camera calibrating, it introduces one kind of camera model, in which has eleven parameters. We take six control points whose space coordinate and image coordinate have been known into the camera model, so we get a set of equations composing eleven unknowns, then we can find its solution. Now we realized the camera calibrating and the image collection so far.Handle to the image that saved in at memory. First, the target object need to makes up a template that should be as small as possible, the template is ok if it can contain principal character of the target thing. Then, through template matching method to ascertain location of the target object in Real-time image, and make note of the image coordinate of target object central point, and the origin of image coordinate system sets up the lower left corner. Finally, display the target thing location area to controlling interface. So far we ascertain the image coordinate of target thing central point.Implement controlling of the manipulator. Space coordinate of target object is ascertained, so the location that it relative to current location of manipulator is also ascertained, so we get the controlled quantity of the manipulator moving forward of target object location. We control the manipulator to arrive at target object location, grab and place the target object. The location of placing can be set up by the consumer. The design realized control manipulator to grab target object with the method of template matching, control accuracy is higher, the hardware realization is simple, and the cost is lower.更多还原