路面坑槽修补智能化技术研究

发布时间：2018-02-20 19:12

本文关键词： 坑槽修补坑槽识别坑槽边缘检测位置跟踪　出处：《长安大学》2017年硕士论文　论文类型：学位论文

【摘要】：针对目前沥青路面坑槽修补方式尤其是采用修补机械实现坑槽修补过程中,存在的自动化、智能化程度不高、效率低、品质差和操作人员劳动强度大等问题,本文提出了基于机器视觉的坑槽智能化修补方式来解决这一问题。本文综合分析自动化坑槽修补过程中坑槽识别的图像特点,确定了基于机器视觉坑槽识别、边缘检测以及机械手定位的工作原理、关键技术、试验装置的硬件组成和软件系统,以及安装布局形式,提出了视觉检测对象和检测方案。针对坑槽边缘识别的视觉模式识别问题,对比传统基于灰度直方图阈值化选取方法和基于数据库比对的方法,发现传统方法鲁棒性和通用性较差。为了更加精确地获取坑槽边缘坐标,提出了用于初期未联结坑槽的边缘检测方法。首先,采用多尺度局部归一化算法对图像进行阴影抑制,再提取限定区域内相似度较高的区域纹理作为样本进行在线的SVM训练,并对坑槽图像进行分类识别预分割及边缘提取。最后基于预分割及边缘提取结果,采用八邻域Prewitt算子对坑槽边缘进行二次分割提取。对于修补机械手的运动定位,采用基于运动控制卡的运动控制方式,基于视觉对机械手进行实时位置识别跟踪反馈,以检测修补程度,为后期交互界面做准备。机械手的位置跟踪包括水平位置和竖直位置的跟踪:水平位置采用基于形状的模板匹配方法,竖直位置采用多平面图像模型进行位置估计。最终试验验证了运动控制方式和机械手的位置跟踪算法的有效性。
[Abstract]:In view of the problems of automation, low intelligence, low efficiency, poor quality and labor intensity of operators in the repair of asphalt pavement potholes, especially in the process of using repair machinery to realize the repair of potholes, there are many problems, such as automation, low intelligence, low efficiency, poor quality and great labor intensity of operators, etc. In this paper, an intelligent repairing method based on machine vision is put forward to solve this problem. In this paper, the image characteristics of pit slot recognition in the process of automatic pit repair are analyzed, and the recognition of potholes based on machine vision is determined. The working principle of edge detection and manipulator positioning, the key technology, the hardware composition and software system of the test device, and the form of installation and layout, Aiming at the problem of visual pattern recognition in pit edge recognition, this paper compares the traditional methods based on grayscale histogram and database comparison. It is found that the traditional methods are less robust and universal. In order to obtain the edge coordinates of the potholes more accurately, an edge detection method for initial unconnected pits is proposed. Firstly, a multi-scale local normalization algorithm is used to suppress the shadow of the image. Then the texture of the region with high similarity in the limited region is extracted as a sample for online SVM training, and the pothole image is classified to recognize the pre-segmentation and edge extraction. Finally, based on the pre-segmentation and edge extraction results, Eight neighborhood Prewitt operators are used to extract the edge of the pit. For the motion location of the mending manipulator, the motion control method based on the motion control card is adopted, and the real-time position recognition and tracking feedback is carried out based on the vision. Position tracking of manipulator includes horizontal position tracking and vertical position tracking: the horizontal position adopts a form-based template matching method. The vertical position is estimated by a multi-plane image model. Finally, the effectiveness of the motion control method and the position tracking algorithm of the manipulator is verified by the experiment.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U418.6;TP391.41

【参考文献】