多线索融合的结构化道路检测算法研究  

Research on Structured Road Detection Algorithm Based on Multi-Cue Fusion

作　　者：李晓威 袁春 李杨 郭宗环 李昊 王瑞虎 LI Xiaowei;YUAN Chun;LI Yang;GUO Zonghuan;LI Hao;WANG Ruihu(Chongqing University of Technology,Chongqing 400054,China;Chongqing Jinkang Seres New Energy Vehicle Design Institute Co.,Ltd.,Chongqing 401133,China;Chongqing Transportation Vocational College,Chongqing 402247,China)

机构地区：[1]重庆理工大学,重庆400054 [2]重庆金康赛力斯新能源汽车设计院有限公司,重庆401133 [3]重庆交通职业学院,重庆402247

出　　处：《汽车工程学报》2023年第5期668-675,共8页Chinese Journal of Automotive Engineering

基　　金：重庆市技术创新与应用发展专项重点项目(cstc2019jscx-mbdX0052)“L4级自动驾驶技术研发”。

摘　　要：为实现结构化道路的检测,提出了一种用于道路检测的激光雷达和视觉融合法,通过提取激光雷达在道路边缘的三维点云信息,将其投影到视觉图像上,形成激光点云与图像的映射关系,生成激光雷达似然图。通过改进提取道路的颜色、纹理、水平线等特征的方法,生成相对应的视觉似然图。在贝叶斯框架下将激光雷达和视觉生成的似然图进行融合。在KITTI数据集上测试可知,精度达到94%,准确率达到86%,表明该道路检测法具有较好的道路检测效果。This paper proposes a method using lidar and vision fusion to achieve structured road detection.The paper extracts the 3D point cloud data from the lidar targeting the edge of the road and projects it onto the visual image to form laser points.The mapping relationship between the laser point cloud and the image is established to generate a lidar likelihood map.The corresponding visual likelihood map is produced by improving the method of extracting features such as color,texture and horizontal lines of the road.Finally,we innovatively fuse the likelihood maps generated from the lidar and visual data within the Bayesian framework.Evaluations on the KITTI dataset show an accuracy of 94% and a success rate of 86%,proving the effectiveness of the proposed road detection method.

关 键 词：激光点云 消失点 似然图融合 贝叶斯框架 

分 类 号：U469[机械工程—车辆工程]