基于模板匹配的视觉分拣方法及应用研究

发布时间：2018-08-15 16:02

【摘要】：随着智能化装备在工业生产中的广泛应用,机器视觉作为智能化装备的一个重要分支受到很大重视。而一直以来国产视觉处理软件在速度、抗干扰等方面性能表现不佳,尽管很多国内企业和研究机构在做相关方面的研究,但国内市场一直被国外的优秀视觉解决方案企业占领很大的份额。本文研究了一种稳定可靠并且高效的视觉识别和定位算法,成功解决多项工程任务,具体开展了如下几个方面的工作:首先,研究了一种基于形状的模板匹配算法,相似度量采用模板边缘的梯度向量和目标物体边缘梯度向量作点积来度量。通过采取贪心算法较大地提高的算法的匹配速度,然后通过图像金字塔降低图像分辨率的方式,降低了算法的复杂度。通过两种速度提升方式,成功地提高了算法匹配速度。并且研究了其它的模板匹配算法,包括基于互相关系数法的匹配方法和性能表现优异的基于豪斯多夫距离和霍夫变换的模板匹配方法。其次,针对不同复杂程度的目标物体,提出了自动确定其图像边缘幅值分割的最低和最高阈值、图像金字塔的层数和模板搜索的平移步长和旋转步长的算法,使得匹配算法更加智能。匹配过程的实现中通过对比试验,确定了图像金字塔建立所需的最佳的滤波方式,解决了最高层金字塔寻优的问题。采用C++编程语言在VS 2013开发环境下开发了视觉识别和定位软件。随后,通过设计实验来验证算法的性能。分为两类实验分别测试算法的抗干扰性能和算法识别所用的时间两项性能。抗干扰性能测试包括抗遮挡、混乱和抗非线性光照变化的性能。本文算法通过与互相关系数法和基于豪斯多夫距离以及基于霍夫变换的三种匹配算法进行对比,验证了本文基于形状的模板匹配算法在处理物体部分信息缺失以及抗光线干扰和处理时间方面的优异性能。最后,搭建了Delta机器人视觉分拣系统,完成了相机与传送带之间的标定,机器人与传送带之间的位置标定,保证分拣系统抓取的准确性。通过基于图像序列的动态目标跟踪算法避免了目标的多次识别和抓取。通过自主开发的视觉处理算法,成功解决了火腿肠自动分拣项目中肠体的识别和定位,并且此项目已经成功应用于实际生产中,达到了高效的分拣任务,大大提高了生产效率。
[Abstract]:With the wide application of intelligent equipment in industrial production, machine vision as an important branch of intelligent equipment has received great attention. And domestic visual processing software has been performing poorly in terms of speed, anti-jamming, and so on, although many domestic enterprises and research institutions are doing related research. But the domestic market has been foreign excellent visual solutions enterprises occupy a large share. In this paper, a stable, reliable and efficient visual recognition and localization algorithm is studied, which successfully solves many engineering tasks. The following works are carried out: firstly, a form-based template matching algorithm is studied. Similarity measures are measured by dot product of gradient vector of template edge and gradient vector of object edge. By adopting greedy algorithm to improve the matching speed of the algorithm, and then reducing the image resolution through the image pyramid, the complexity of the algorithm is reduced. The matching speed of the algorithm is improved successfully by two speed lifting methods. And other template matching algorithms are studied, including the matching method based on the correlation number method and the template matching method based on the Hausdorf distance and Hoff transform with excellent performance. Secondly, an algorithm is proposed to automatically determine the minimum and highest threshold values of image edge amplitude segmentation, the number of layers of image pyramid and the translation step size and rotation step size of template search for different complex objects. It makes the matching algorithm more intelligent. In the realization of the matching process, the optimal filtering method for image pyramid building is determined through the contrast experiment, and the problem of the top pyramid optimization is solved. Visual recognition and location software is developed by using C programming language in vs 2013 development environment. Then, the performance of the algorithm is verified by designing experiments. It is divided into two kinds of experiments to test the performance of anti-jamming and the time of recognition respectively. Anti-jamming performance tests include anti-occlusion, chaos, and anti-nonlinear illumination variations. The algorithm is compared with the correlation number method, the three matching algorithms based on Hausdorf distance and Hoff transform. It is verified that the shape based template matching algorithm has excellent performance in dealing with the missing information of objects and the ability to resist light interference and processing time. Finally, the Delta robot visual sorting system is built, the calibration between camera and conveyor belt is completed, and the position calibration between robot and conveyor belt is completed to ensure the accuracy of the grabbing system. The dynamic target tracking algorithm based on image sequence is used to avoid multiple target recognition and capture. Through the self-developed visual processing algorithm, the identification and location of the midgut body of the ham sausage automatic sorting project has been successfully solved, and this project has been successfully applied in practical production, which has achieved a highly efficient sorting task and greatly improved the production efficiency.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】