双目视觉下的自动驾驶车辆多类型障碍物感知  

Perception of multiple types of obstacles in autonomous vehicles under binocular vision

作　　者：郭迎 赵祥模[2] 梁睿琳 王润民[2] GUO Ying;ZHAO Xiangmo;LIANG Ruilin;WANG Runmin(The Joint Laboratory for Internet of Vehicles,Ministry of Education-China Mobile Communications Corporation,Chang′an University,Xi′an 710018,China;School of Information Engineering,Chang′an University,Xi′an 710018,China)

机构地区：[1]长安大学车联网教育部-中国移动联合实验室,陕西西安710018 [2]长安大学信息工程学院,陕西西安710018

出　　处：《哈尔滨工程大学学报》2023年第8期1406-1411,共6页Journal of Harbin Engineering University

基　　金：国家重点研发计划(2021YFB2501200);陕西省重点研发计划(2021LLRH-04-01-03);中央高校基本科研业务费(300102242501).

摘　　要：为解决视差变化率与距离变化率呈现非线性关系而导致障碍物感知误差大,本文提出基于双目视觉的多类型障碍物感知方法。利用极性几何约束双目摄像机相对位置关系,将多视极线约束与双目视觉下位置运动信息相结合,高斯滤波滤除图像噪声。采用卷积函数算子提取图像特征信号的角点,设置查找窗口,匹配左右目视觉下图像的相似程度,提高像素点集正确匹配准确率。研究表明:所提方法能够准确感知不同类型障碍物真实位置,障碍物特征匹配正确率高,实际感知误差小,该方法可为智能交通和城市交通管理提供可靠的数据支持,减少交通事故的发生率。A multitype obstacle perception method based on binocular vision is proposed to solve the problem of large obstacle perception errors due to the nonlinear relationship between parallax and distance change rates.The multiview epipolar constraint is combined with the position motion information under binocular vision based on the relative position relationship of binocular cameras constrained by polar geometry,and the image noise is sorted by a Gaussian filter.The convolution function operator is used to extract the corners of the image feature signals,and the search window is set to match the image similarity under the left and right visions to improve the accuracy of the correct matching of the pixel set.Experiment results prove that the proposed method can accurately perceive the real positions of different types of obstacles while demonstrating high accuracy in obstacle feature matching and a small actual perception error.In summary,this method can provide reliable data support for intelligent transportation and urban traffic management,thereby reducing the incidence of traffic accidents.

关 键 词：障碍物检测 双目视觉 场景变换 矩阵约束 特征匹配 无人驾驶 极性几何约束 多视极线约束 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]