结合立体视觉的非结构化道路消失点检测研究  

Research on vanishing point detection of unstructured road scene combined with stereo vision

作　　者：陈星 张文海 杨林 郑讯佳 CHEN Xing;ZHANG Wenhai;YANG Lin;ZHENG Xunjia(School of Intelligent Manufacturing,Chongqing University of Arts and Sciences,Chongqing 402160,China;School of Mechatronics and Vehicle Engineering,Chongqing Jiaotong University,Chongqing 400074,China;Special Vehicle Research Institute,Chongqing Changan Industry(Group)Co.,Ltd,Chongqing 400023,China)

机构地区：[1]重庆文理学院智能制造工程学院,重庆402160 [2]重庆交通大学机电与汽车工程学院,重庆400074 [3]重庆长安工业(集团)有限责任公司特种车辆研究所,重庆400023

出　　处：《西北工业大学学报》2022年第6期1431-1439,共9页Journal of Northwestern Polytechnical University

基　　金：重庆市教委科学技术研究项目(KJZD-K202101301);中国博士后科学基金(2019T120813,2018M643420);重庆文理学院人才引进项目(R2021SZZ01)资助。

摘　　要：消失点检测是基于视觉的无人车辆自主导航的重要组成部分。由于非结构化场景存在缺乏清晰的道路线和复杂的背景干扰等问题,现有检测方法普遍存在精度低、计算时间长的缺点。因此,针对非结构化道路特点,提出了一种结合立体视觉的消失点检测方法。采用双目立体视觉技术获得道路图像的视差图,使用广度优先算法快速估计出道路图像的背景区域;设计四方向五尺度的Gabor滤波器组估计像素响应幅值,并通过幅值校正减少检测误差;结合背景区域设计一系列投票点选择策略,来剔除背景区域的干扰,提高算法精度;采用动态调整候选点范围策略,减少消失点的搜索范围,从而提高算法效率;设计了一种角度优先投票函数,将在投票空间中获得最大票数的候选点视为消失点。结果表明,改进的方法在复杂背景干扰的场景下具有较好的鲁棒性,在检测速度和检测精度上都有显著提升。Vanishing point detection is an important part of visual-based autonomous navigation of unmanned vehicles. Due to the lack of clear routes and complex background interference in unstructured road scenes, existing detection methods generally have the shortcomings of low accuracy and long calculation time. Therefore, according to the characteristics of unstructured road, a vanishing point detection method combined with stereo vision is proposed in this paper. Binocular stereo vision technology is used to obtain the parallax image of road image, and breadth-first algorithm is applied to the parallax image to estimate the background area of road image quickly. Gabor filter banks with 4 directions and 5 scales are designed to estimate the pixel response amplitude, and the detection error is reduced by amplitude correction. Combined with the background region, a series of voting place selection strategies are designed to eliminate the interference of the background region and improve the accuracy of the algorithm. A strategy of dynamically adjusting the range of candidate points is proposed to reduce the search range of vanishing points to improve the efficiency of the algorithm. Finally, an angle preference voting function is designed, which considers the candidate with the largest number of votes in the voting space as the vanishing point. Simulation results show that the improved method has better robustness in the scene with complex background interference, and significantly improves the detection speed and accuracy of vanishing point detection.

关 键 词：消失点检测 立体视觉 广度优先 GABOR滤波器组 动态调整 背景干扰 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]