摘要
针对倒车过程中双目视觉图像障碍物检测时的误匹配、障碍物检测范围等问题,提出基于双目视觉图像的倒车障碍物检测方法。利用膨胀和腐蚀操作来消除误匹配及部分噪声;采用视差图像的二值化处理来控制障碍物的检测范围,实现基于距离的障碍物检测;接着使用基于行程的连通区域标记算法对各连通区域进行标记;最后计算各区域面积,将面积过小的区域作为噪点去除,将剩余区域作为障碍物提出来。实验结果表明笔者提出的基于双目视觉图像的倒车障碍物检测方法能够有效检测出车后图像的障碍物区域。
According to the problem of false matching and obstacle detection range of binocular vision image obstacle detection in the course of reversing,a reversing obstacle detection method based on binocular stereo vision was proposed.Firstly,dilation and erosion algorithms were implemented in the disparity map to connect discontinuous region and remove small noise points.Binaryzation processing of parallax images was adopted to control the detection range of obstacles,which realized the detection of obstacles based on distance.Then the connectivity region marking algorithm based on travel was used to mark all connected regions.Finally,the area of the above regions was calculated.The region with too small area was removed as hot pixel,and the remaining area was proposed as an obstacle.The experiment results demonstrate that the proposed detection method of reversing obstacle based on binocular vision image can effectively detect the obstacle area of the post-car image.
作者
刘昱岗
王卓君
刘艳芳
张祖涛
徐宏
LIU Yugang1,2 , WANG Zhuojun1, LIU Yanfang1 , ZHANG Zutao3 , XU Hong4(1.School of Transportation and Logistics, Southwest Jiaotong University, Chengdu 610031, Sichuan, P. R. China; 2. National United Engineering Laboratory of Integrated and Intelligent Transportation, Southwest Jiaotong University, Chengdu 610031, Sichuan, P. R. China; 3. School of Mechanical Engineering, Southwest Jiaotong University, Chengdu 610031, Sichuan, P. R. China; 4. Thirty-second Research Institute of China Electronic Science and Technology Group, Shanghai 201808, P. R. Chin)
出处
《重庆交通大学学报(自然科学版)》
CAS
北大核心
2018年第3期92-98,共7页
Journal of Chongqing Jiaotong University(Natural Science)
关键词
交通运输工程
智能交通
双目视觉
倒车
区域标记
障碍物检测
traffic and transportation engineering
intelligent transportation
binocular vision
reversing
region labelling
obstacle detection