文献类型：会议论文

英文题名：Robust Free-Space Detection in Urban Roads based on MSER Extraction Using Gradient Images

作者：Xin, Le[1,2,3];Song, Juan[1,4];Chen, Yangzhou[1,2,3];Hu, Jiangbi[1,2,3,5]

第一作者：Xin, Le

通讯作者：Xin, L[1];Xin, L[2];Xin, L[3]

机构：[1]Beijing Univ Technol, Coll Metropolitan Transportat, Beijing 100124, Peoples R China;[2]Beijing Key Lab Traff Engn, Beijing 100124, Peoples R China;[3]Beijing Engn Res Ctr Urban Transport Operat Guara, Beijing 100124, Peoples R China;[4]Beijing Union Univ, Coll Automat, Beijing 100101, Peoples R China;[5]Beijing Univ Technol, Coll Architecture & Civil Engn, Beijing 100124, Peoples R China

第一机构：Beijing Univ Technol, Coll Metropolitan Transportat, Beijing 100124, Peoples R China

通讯机构：[1]corresponding author), Beijing Univ Technol, Coll Metropolitan Transportat, Beijing 100124, Peoples R China;[2]corresponding author), Beijing Key Lab Traff Engn, Beijing 100124, Peoples R China;[3]corresponding author), Beijing Engn Res Ctr Urban Transport Operat Guara, Beijing 100124, Peoples R China.

会议论文集：37th Chinese Control Conference (CCC)

会议日期：JUL 25-27, 2018

会议地点：Wuhan, PEOPLES R CHINA

语种：英文

外文关键词：Road Detection; On-board Video Processing; Free-Space Detection; MSER; Gradient Images

摘要：In this paper, a robust algorithm to detect free-spaces in urban roads using video sequences captured on-board is proposed, mainly aiming at the complex urban road conditions such as the change of road structure and dense traffic conditions. as well as the changing illumination. This algorithm is based on MSER (Maximally Stable Extremal Region) extraction using gradient images. which includes the following two aspects. Firstly. the gradient images are calculated by the Sobel operator formed as the spatial filter template. MSERs are extracted from these gradient images with an improved MSER extraction algorithm with linear complexity. Then some shape analysis is applied on each MSER after shape representation is computed by the contour extraction. On the basis of selecting the proper pixel location of the initial seed point, the region of the road area is determined where the initial seed point falls, based on the assumption that the region directly in front of the host vehicle is just the road area. Experimental results show that this method can run well in various traffic scenarios, and achieve real-time and robust road detection results.