Unstructured Road Detection Based On Monocular Vision

Road driving area identification is an important part of intelligent vehicle vision technology,and is often used in mobile robot systems,intelligent vehicle assisted driving and other fields.This paper studies the method of detecting the area of the intelligent vehicle on the unstructured road on a single frame image.The main contents include three aspects: road vanishing point detection,vertical object detection and road travelable area segmentation.First of all,for the upper part of the image,the unrelated information such as sky and clouds will affect the image texture detection and RGB entropy segmentation of the road surface.In this paper,the Gabor texture is used to detect the vanishing point of the road,and the unwanted information above the vanishing point is removed to reduce the interference.The method first uses the Gabor texture detector to calculate the texture information in the image;then determines whether the point is used as a voting point according to the texture confidence criterion;finally,the FLASV voting method is used to screen out the vanishing point of the road.The experimental part proves the feasibility of the vanishing point detection algorithm.Secondly,according to the fact that the dynamic targets such as passers-by and vehicles on the plane of the road can be approximated as perpendicular to the columnar objects on the road surface,this paper proposes a vertical object detection method based on depth information.The method constructs a Gaussian template by using the disparity value,and projects the vertical object onto the place-occupyinggrid with a certain probability of occupying,and then uses the dynamic programming method to find the best obstacle grounding point.The height of the final object is calculated by the V-disparity map theory.The detected vertical object is projected onto the original image.The experimental analysis partially proves the feasibility of the algorithm.Finally,in order to overcome the problem that RGB entropy can't distinguish the background and road with similar colors,the road edge and texture features are introduced as constraints to modify the RGB entropy segmented road region to adapt to the similarity between the road region and the background region.Finally,the experimental analysis verifies the feasibility of the proposed algorithm and can detect the road boundary more accurately.It is also applicable to curved road surfaces.