移动机器人图像处理关键技术研究与实现
发布时间：2019-03-08 18:06
【摘要】：移动机器人视觉导航过程中图像处理的关键问题是道路识别和障碍物检测,论文是基于计算机单目视觉技术对非结构化道路识别和运动障碍物检测进行研究。在已有的技术基础上,经分析和实验,本文采用彩色道路边缘检测并结合道路区域识别技术,对校园环境下的非结构化道路识别能够取得较好的结果,算法对复杂的野外道路环境也有一定的识别效果;对于场景中障碍物检测技术,本文采用三帧差分法结合金字塔光流法,能够对道路中的运动目标进行检测,判定其在图像中的位置。本文研究内容主要分为四个方面:移动机器人软件系统设计、图像道路区域识别、道路边缘检测、障碍物检测。移动机器人软件系统设计方面,本文主要研究的是移动机器人视觉导航系统的设计。视觉导航系统的作用是:把从传感器获取到的场景信息流经过系统中道路识别、障碍物检测、运动决策等模块的综合处理来完成导航任务。视觉导航系统的设计是通过模块化、多线程等方法来完成,以便系统具有较好的可维护性和实时性。道路区域识别方面,本文回顾了道路识别常用算法类型,详细介绍了基于区域的道路识别,本文使用颜色特征结合LBP纹理特征作为图像总特征,采用监督的训练方式,使用二分类效果优良的SVM算法训练分类器。用训练好的分类器对图像道路区域进行粗识别。在道路边缘检测方面,本文分析了5种常用的边缘检测算子在道路识别中的效果,并结合实验分析,采用了一种彩色空间下改进的Canny边缘检测算法。改进的Canny边缘检测,不同于传统算法在2x2的邻域内计算两个方向均值差分,本文采用3x3的4个方向的部分插值,同时由于移动机器人场景的不断变化,本文没有使用传统方法的固定阈值,而是根据每幅图像采用最大类间方差的自适应阈值。改进后的Canny边缘检测算法更适合移动机器人自主导航场景。在障碍物检测检测方面,本文主要研究移动机器人导航过程中的运动障碍物的检测。本文详细的介绍了障碍物检测常用的三种算法:背景模型差分法、两帧差分法、光流法。然后依据移动机器人的硬件资源,在满足实时和准确性的前提下,使用一种基于三帧差分结合金字塔光流的障碍物检测算法,能实时的检测到移动机器人前方场景中的运动目标。
[Abstract]:The key problem of image processing in vision navigation of mobile robot is road recognition and obstacle detection. This paper studies unstructured road recognition and moving obstacle detection based on computer monocular vision technology. On the basis of existing technology, through analysis and experiment, this paper adopts color road edge detection combined with road area recognition technology, which can achieve better results for unstructured road recognition in campus environment. The algorithm also has a certain recognition effect on the complicated road environment in the field. In this paper, the three-frame difference method combined with pyramid optical flow method is used to detect the moving object in the road and determine its position in the image. The main contents of this paper are divided into four aspects: mobile robot software system design, image road area recognition, road edge detection, obstacle detection. In the design of mobile robot software system, this paper mainly focuses on the design of vision navigation system of mobile robot. The function of the visual navigation system is to complete the navigation task by integrating the information flow of the scene obtained from the sensor through the integrated processing of the system modules such as road recognition, obstacle detection, motion decision-making and so on. The design of visual navigation system is accomplished by modularization and multithreading, so that the system has better maintainability and real-time. In the aspect of road area recognition, this paper reviews the common algorithm types of road recognition, and introduces the road recognition based on region in detail. In this paper, the color feature combined with LBP texture feature is used as the total image feature, and the supervised training method is adopted. The classifier is trained by using the SVM algorithm, which has a good effect of binary classification. The trained classifier is used to rough recognize the road area of the image. In the aspect of road edge detection, this paper analyzes the effect of five commonly used edge detection operators in road recognition, and adopts an improved Canny edge detection algorithm in color space combined with experimental analysis. The improved Canny edge detection is different from the traditional algorithm in calculating the mean difference between two directions in the neighborhood of 2x2. In this paper, we use the partial interpolation of 3x3 in four directions, at the same time, due to the continuous change of the mobile robot scene, In this paper, we do not use the fixed threshold of the traditional method, but adopt the adaptive threshold of the maximum inter-class variance according to each image. The improved Canny edge detection algorithm is more suitable for autonomous navigation of mobile robot. In the aspect of obstacle detection, this paper mainly studies the moving obstacle detection in the process of mobile robot navigation. In this paper, three common algorithms for obstacle detection are introduced in detail: background model difference method, two frame difference method and optical flow method. Then according to the hardware resources of the mobile robot, an obstacle detection algorithm based on three-frame difference and pyramid optical flow is used on the premise of satisfying the real-time and accuracy. It can detect the moving objects in front of the mobile robot in real time.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242
[Abstract]:The key problem of image processing in vision navigation of mobile robot is road recognition and obstacle detection. This paper studies unstructured road recognition and moving obstacle detection based on computer monocular vision technology. On the basis of existing technology, through analysis and experiment, this paper adopts color road edge detection combined with road area recognition technology, which can achieve better results for unstructured road recognition in campus environment. The algorithm also has a certain recognition effect on the complicated road environment in the field. In this paper, the three-frame difference method combined with pyramid optical flow method is used to detect the moving object in the road and determine its position in the image. The main contents of this paper are divided into four aspects: mobile robot software system design, image road area recognition, road edge detection, obstacle detection. In the design of mobile robot software system, this paper mainly focuses on the design of vision navigation system of mobile robot. The function of the visual navigation system is to complete the navigation task by integrating the information flow of the scene obtained from the sensor through the integrated processing of the system modules such as road recognition, obstacle detection, motion decision-making and so on. The design of visual navigation system is accomplished by modularization and multithreading, so that the system has better maintainability and real-time. In the aspect of road area recognition, this paper reviews the common algorithm types of road recognition, and introduces the road recognition based on region in detail. In this paper, the color feature combined with LBP texture feature is used as the total image feature, and the supervised training method is adopted. The classifier is trained by using the SVM algorithm, which has a good effect of binary classification. The trained classifier is used to rough recognize the road area of the image. In the aspect of road edge detection, this paper analyzes the effect of five commonly used edge detection operators in road recognition, and adopts an improved Canny edge detection algorithm in color space combined with experimental analysis. The improved Canny edge detection is different from the traditional algorithm in calculating the mean difference between two directions in the neighborhood of 2x2. In this paper, we use the partial interpolation of 3x3 in four directions, at the same time, due to the continuous change of the mobile robot scene, In this paper, we do not use the fixed threshold of the traditional method, but adopt the adaptive threshold of the maximum inter-class variance according to each image. The improved Canny edge detection algorithm is more suitable for autonomous navigation of mobile robot. In the aspect of obstacle detection, this paper mainly studies the moving obstacle detection in the process of mobile robot navigation. In this paper, three common algorithms for obstacle detection are introduced in detail: background model difference method, two frame difference method and optical flow method. Then according to the hardware resources of the mobile robot, an obstacle detection algorithm based on three-frame difference and pyramid optical flow is used on the premise of satisfying the real-time and accuracy. It can detect the moving objects in front of the mobile robot in real time.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242
【参考文献】
相关期刊论文 前8条
1 武怀金;王武江;;基于帧间差分方法的运动目标检测[J];黑龙江科技信息;2012年10期
2 刘威;于红绯;杨恒;段勃勃;袁淮;赵宏;;一种新的基于单目视觉的广义障碍物检测方法[J];电子学报;2011年08期
3 薛丽霞;罗艳丽;王佐成;;基于帧间差分的自适应运动目标检测方法[J];计算机应用研究;2011年04期
4 杨s，
本文编号：2437071
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2437071.html
