基于立体视觉的野外环境坑区域识别研究

CORC > 沈阳自动化研究所 > 中国科学院沈阳自动化研究所 > 机器人学研究室

题名	基于立体视觉的野外环境坑区域识别研究
作者	孟令江
学位类别	硕士
答辩日期	2016-05-25
授予单位	中国科学院沈阳自动化研究所
导师	王挺
关键词	机器视觉阈值分割坑区域圆相似性立体匹配
其他题名	Research on Wild Pit Region Detection Based on Stereo Vision
学位专业	控制工程
中文摘要	野外环境中有很多坑区域，这给工作在此环境中机器人的移动带来很大的困难。传统的对环境的检测方法有红外、超声波、激光和视觉等，本文使用视觉来实现对坑区域的检测。首先通过标定获取左右相机矫正后的图像，之后对图像进行预处理，然后通过阈值分割和非规则圆区域的提取获取疑似坑区域，最后利用双目视觉进一步实现坑的判定并完成坑区域的定位和测距。具体工作如下：1.使用Opencv标定程序与Matlab标定工具箱配合完成摄像机标定。2.对图像进行了滤波和形态学处理；根据工程需要和野外坑特点去除一部分不满足要求区域。为了确定阈值范围，使用了已有的两种经典阈值分割方法和本文设计的两种算法来对图像进行分割试探，为下一步提取疑似坑区域做好准备。3.分析了已有的有关圆提取算法在面对野外坑区域时具有的不足，分析了野外环境下坑区域所具有的特点，本文设计了两种结合坑特征的阈值分割方法。通过阈值面积消去、计算梯度、区域旋转、圆相似度计算、计算扁平率与出现频率之间关系等步骤有效提取出了疑似坑区域。4.根据野外环境下坑与周围环境的特点。设计了SURF特征点匹配和半全局匹配结合使用的方案，完成了疑似坑区域的判定、测距和定位。通过实验最后验证了本文所使用的方法和方案提取准确率高、鲁棒性好，运行时间短，是一种有效的方法。本文的工作丰富了机器视觉在野外环境中有关坑提取的研究内容，对工作于野外环境中机器人的视觉检测处理提供了一种解决方案。
英文摘要	There are many pit regions in the wild environment, which brings difficulties for robot movement in the circumstances. The traditional methods to identify the environment include infrared method, ultrasonic method, laser method and vision method. This paper uses the method of machine vision to detect pit region. Firstly, get the rectified images after camera calibration; then get the pit region with threshold segmentation and irregular circle region extraction; after image preprocessing, the pit region acquired; lastly, pit region can be determined by using binocular vision. Thus, the location and distance measurement can be completed. The specific studies are as follows: 1. The Opencv calibration procedure, cooperated with the Matlab calibration kit is used to achieve camera calibration. 2. Images have been processed by using filter and morphology. According to the requirement of project and the characteristics of pit region, many regions which cannot meet the requirement have been removed. In order to acquire the range of threshold, two classical segmentation algorithms and two other algorithms in this paper are used for image segmentation in comparison, which sets the preparation for pit extraction in next processing. 3. The defects of the algorithms existing for wild pit detection are analyzed, and the characteristic of wild pit is analyzed. This paper design two image extraction methods with threshold segmentation based on the characteristic of wild pit. By using threshold area elimination, gradient computation, region rotation, similarity of pit computation, the relation of the aspect ratio computation and occurrence of frequency etc, the suspect pit regions can be extracted. 4. Based on the characteristic of wild pits' and their surrounding environment, the scheme of combination of surf algorithm with semi global matching is designed for achieving determinating pit region, measuring distance and locating. The method and scheme are verified by experiment with high extraction accuracy, high robustness and short running time, so this method and scheme is believed to be effective. The work of this paper enrich the research of pit extraction of machine vision in the wild environment, and present a solution for vision processing used by robot that works in the wild environment.
语种	中文
产权排序	1
页码	75页
内容类型	学位论文
源URL	[http://ir.sia.cn/handle/173321/19675]
专题	沈阳自动化研究所_机器人学研究室
推荐引用方式 GB/T 7714	孟令江. 基于立体视觉的野外环境坑区域识别研究[D]. 中国科学院沈阳自动化研究所. 2016.