智能车辆基于视觉的可通行区域检测方法

发布时间：2018-04-24 03:17

本文选题：垂直投影校正 + 加权投票　；参考：《南京理工大学》2017年硕士论文

【摘要】：可通行区域检测是智能车辆视觉技术的关键技术之一,广泛应用于智能交通系统、车辆辅助驾驶系统等领域。基于单一的图像,本文从消失点、障碍物和道路分割三个方面研究了智能车辆在非结构化道路上的可通行区域检测方法。首先,针对弯曲道路消失点检测不准确的问题,本文提出了一种新的基于加权投票的消失点检测方法。该方法先用垂直投影校正方法来校正投票点的Gabor纹理主方向,再通过投票点对上方的、位于同一条边缘线上的其他投票点进行加权的方法计算出投票权值,最后用该权值进行加权投票,以检测出消失点。消失点检测实验验证了本文方法检测效果最好,并且可以满足系统实时性要求。其次,基于树木、栏杆等障碍物都有一定竖直结构的事实,本文提出了一种新的基于Haar-like特征和纹理主方向的竖直障碍物检测方法。该方法先使用线性矩形特征区分出障碍物与背景,再通过在矩形区域内垂直纹理概率多少来判别该矩形是否包含障碍物。后面的障碍物检测实验表明,该方法可以准确检测出竖直障碍物,并且障碍物矩形内投票点进行逆投票,可以消除障碍物对消失点检测的干扰。最后,为了提高弯曲道路分割的准确度,本文提出了基于消失点的分段道路分割方法。该方法先将图像在消失点以下的区域按高度比例分为两部分,再基于真实道路在图像中的形状情况,一一计算边界。道路分割实验验证了本文方法满足系统实时性的要求,并且提高了道路分割的准确度。
[Abstract]:The detection of passable area is one of the key technologies of intelligent vehicle vision technology. It is widely used in the fields of intelligent transportation system and vehicle assisted driving system. Based on a single image, this paper studies the method of detecting the passable area of intelligent vehicle on unstructured road from three aspects: vanishing point, obstacle and road segmentation. Firstly, a new vanishing point detection method based on weighted voting is proposed to solve the problem of inaccurate detection of vanishing points in curved roads. This method uses the vertical projection correction method to correct the main direction of the Gabor texture of the polling point, and then calculates the voting value by weighting the other polling points located on the same edge line above the polling point. Finally, weighted voting with the weight value is used to detect the vanishing point. The experiment of vanishing point detection shows that the proposed method has the best detection effect and can meet the real-time requirement of the system. Secondly, based on the fact that obstacles such as trees and railings have a certain vertical structure, a new method of vertical obstacle detection based on Haar-like features and texture principal direction is proposed in this paper. The method first uses linear rectangular features to distinguish obstacles from backgrounds, and then determines whether the rectangle contains obstacles by the probability of vertical texture in the rectangular region. The experiment of obstacle detection behind shows that the method can accurately detect vertical obstacles and reverse vote at polling points in the rectangle of obstacles, which can eliminate the interference of obstacles to detection of vanishing points. Finally, in order to improve the accuracy of curved road segmentation, a segmented road segmentation method based on vanishing points is proposed. In this method, the region below the vanishing point is divided into two parts according to the height, and then the boundary is calculated one by one based on the shape of the real road in the image. Road segmentation experiments show that the proposed method meets the real-time requirements of the system and improves the accuracy of road segmentation.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;TP391.41

【相似文献】