| An efficient and accurate detection of the road region is one of the most important technologies in the vehicle navigation part in an Advanced Driver Assistant System.For the past few years,with the development of the performance of computers and cameras,road detection technologies based on vision information achieve a rapid progress.In this paper,we carry on a research on the road detection technology based on Computer Vision.Our work is divided into three parts:road detection based on monocular vision,road detection based on stereo vision and road detection based on deep learning.We provide different methods in each part respectively,to help solve the problems of the segmentation of road region.The details are listed as follows:(1)In this part,we try to recognize the road region from images taken by monocular cameras.For a common outdoor image,an effective preprocessing means a lot to the following object recognition and scene segmentation.After a shadow removal scheme,we significantly decrease the effect of shadows on the road.With the accurate detection of vanishing point,combining with the edge information and image segmentation,we get a preferable road segmentation result.(2)Stereo vision is utilized in this part.With the depth information computed from binocular images,we can significantly improve the performance of road segmentation.We need to calibrate the binocular camera system accurately from the left and right images we have got,and match the corresponding features of each image.With the disparity information from the left and right images,we can get the depth information of the road surface and the sidewalk,pedestrians and other obstacles,and use it to get the road region we want.(3)Based on the deep convolutional network,we construct a network which is suitable for the task of urban road detection,named as convolutional patch network.We train our network using the open dataset KITTI to have optimal parameters and consequently achieve a decent road segmentation result. |
PDF Full Text Request