【作者】 王彦；

【导师】 傅卫平；

【作者基本信息】 西安理工大学 ， 机械工程， 2010， 博士

【摘要】 机器视觉作为获得环境信息的主要手段之一,可以增加工业机器人的自主能力,提高其灵活性。工业机器人通过视觉如何根据所获取的图像信息,正确地、实时地提取出工件特征参数、识别出工件类型并判断出工件所处的位置姿态,是机器视觉应用于工业领域的关键技术之一。本文以机械手目标识别和定位算法为研究目标,主要针对工件的实时匹配识别技术和空间定位技术进行了详细的研究。主要工作如下：1.为了解决亮度不均匀等复杂环境下的图像匹配问题,改进了一种基于边缘匹配的工件识别算法。该算法对原始图和模板图采用样条小波进行增强,用Canny算子提取的边缘信息作为匹配特征,将改进的Hausdorff距离作为图像匹配的相似性度量,在搜索过程中采用了基于种群代沟信息的自适应遗传算法,在不损失解的质量的情况下,使遗传算法求解效率得到明显的改善。实验结果表明,该算法不仅加快了匹配过程,对于光照条件的变化具有很好的抵抗能力,而且能有效解决不易提取边缘信息等情况下的图像匹配识别问题。2.为了解决平移、旋转、缩放和部分遮挡等复杂环境下的工件图像匹配问题,给出了一种基于SIFT特征匹配的工件识别算法。该算法采用SIFT(尺度不变特征变换)特征作为匹配特征,引入欧氏距离作为图像匹配的相似性度量,最后对特征点进行错配消除,大大提高了匹配的速度。实验结果证明,该算法能有效解决具有平移、旋转、缩放和部分遮挡等情况下的工件匹配识别问题。3.小波变换具有数据压缩和检测信号局部突变的能力,而SIFT(尺度不变特征变换)对于平移、旋转、缩放和部分遮挡具有不变性。结合小波变换与SIFT特征提出了一种有效的工件图像匹配方法。该方法将原始图和模板图做小波分解以获得粗尺度的平滑图像；利用DoG算子对工件图像进行关键点检测,进而用欧氏距离对关键点进行特征匹配,最后对特征点进行错配消除。因此,两者优势的结合不但可以有效减少工件图像匹配的计算量,而且还可以减弱对于图像采集平台拍摄方位、拍摄距离、角度、光照条件等的依赖性,提高算法的实用性。4.结合生产制造过程中工件的自动定位、识别、计数、分类等问题,为满足硬件及实时性要求,针对DSP6000系列VC4016嵌入式智能相机(以下简称VC4016),提出一种基于视觉的工件特征向量匹配识别方法,对每个工件进行特征提取,并建立向量对应关系,采用特征向量匹配方法与基于神经网络实现工件的识别。然后,应用VC4016,采用色彩转换,RlC(run length code)图像处理技术和Socket通信技术,实现工件自动定位识别系统的软件开发。5.对目标的三维重建工作进行了研究,主要从视觉系统的构建入手,通过对常用的摄像机模型及其标定方法的分析,建立了同时对两摄像机进行标定来推导两摄像机位姿关系的方法。鉴于棋盘格图像应用在本文的摄像机标定中,因此为了对边缘模糊的棋盘格图像进行在线标定,提出了一种改进棋盘格图像角点检测效果的方法,该方法是对标定板图像采用小波增强,以提高标定板图像的清晰度。利用图像点及摄像机的标定参数求解空间点的位置,根据两摄像机之间的位姿关系,把图像点在某一摄像机坐标系下的坐标值转化为另一摄像机坐标系下的坐标值,从而简化了空间点的求解问题。更多还原

【Abstract】 With the rapid developments of computer technology, image processing technology and pattern recognition technology, machine vision is applied more widely. As one important way to obtain environment information, the machine vision can improve robotic flexibilities and autonomy. It is a key technology that how robots with vision to capture exact image information, as well as to extract the feature parameters of components real-timely, to recognize the component types, and to judge the position and posture of component. Based on the reason, the paper presents the algorithms of target recognition and position in industrial robot, especially real-time match recognition technology and space position technology in detail. The main task follows:1. To solve the problem of image matching under complex conditions including background and uneven illumination, a work-piece recognition algorithm based on edge matching app roach is imp roved. The algorithm enhances primitive image and temp late image by spline wavelet, adopts the edge information captured by Canny operator as a matching feature, applies improved Hausdorff distance to measure the degree of similarity between two objects and performs the search by adaptive genetic algorithm based on population generation gap information. The efficiency of the genetic algorithm is imp roved obviously without loss of quality. Experimental results show that the proposed algorithm not only speeds up the matching process greatly and is robust to the illumination variation but also effectively solves the problem of image matching and recognition when the edge information is not easy to be extracted.2. To solve the problem of work-piece image matching under the complex circumstance of translation, rotation, scale and part of occlusion, an algorithm of work-piece recognition based on SIFT (Scale Invariant Feature Transform) is suggested in this paper. The algorithm uses SIFT characteristics as matching features, then introduces the Euclidean distance as the similarity metrics of image matching, and uses a method of setting a threshold value to delete the false matching points. The experimental results proved that the algorithm can effectively solve stereo matching problems of work-piece images including translation, rotation, scale and part of occlusion.3.Wavelet transform provides itself with the capability of data compression and detecting local signal mutation, while scale invariant feature transform (SIFT) have the invariant ability of translation, rotation, scale and part of occlusion. An effective method of work-piece image matching is proposed by combining wavelet transform with SIFT. The smooth image of coarse scale can be obtained through wavelet decomposition for the original image and the model image by this method. The key points are detected using difference of Gaussians (DoG) operator and are matched using Euclidean distance, then the points of error matching are eliminated. Therefore, combination of these two methods can effectively reduce computation cost for work-piece image matching and the dependence on image acquisitive position, direction, light condition etc, which imp roves the algorithm application.4. In order to satisfy the requirements of hardware and real-time performance, the topic offers a vector matching method for part recognition based on vision using the embedded intelligent camera of DSP6000 (hereinafter referred to as VC4016) with the problem in automatic positioning, recognizing, counting and classing etc. in the process of manufacture. Extracting the feature of each part and building the corresponding relation of vector, the work piece can be recognized by making use of the matching method of characteristic number and eigenvector based on higher order neural network. After using color conversion, RIC image processing technology and Socket communication technique, to achieve software development about Automatic recognition and positioning system of work-piece.5. Three-dimensional reconstruction of target, mainly from the construction of visual system to start, has been studied. By analyzing usual camera model and its calibration method,a method was proposed about solving the pose relation between two cameras. X-corner patterns are used in camera calibration. In order to do online calibration for blurred image and improving image definition on calibration board, we proposed a method for X-corner detection based on the special character of X-corners, by using image enhancement based on wavelet analysis on calibration board. Using markers in image coordinates system, calibration parameters of two cameras and the pose relation about two cameras, image points about two cameras in left and right coordinate system can be converted into the same coordinate system, so solving of the spatial point can be simplified.更多还原