基于双目视觉的智能车辆障碍物探测技术研究

Study on Binocular Vision Based Obstacle Detection Technology for Intelligent Vehicle

鉴于障碍物探测是越野智能车辆自主导航的关键环节,为此针对越野环境光照多变、地形复杂的特点,提出了一种适用于越野环境的双目视觉障碍物检测技术,即首先对系统进行标定和坐标变换,以抵消地形的影响;然后采用高斯滤波和有限对比适应性直方均衡化(CLAHE)对图像进行预处理,以削弱噪声、光照和对比度的影响;接着在特征匹配部分,用提取的图像的亚像素级Harris角点特征参与匹配;同时基于RANSAC方法估计基础矩阵,再通过对极几何约束匹配来提高系统的实时性,并采用连续性约束消除误匹配,最终获取环境的3维信息;在障碍物提取部分,则通过线性插值来构建车前环境的高程图像;最后通过边缘提取和形态学处理来最终检测障碍物。此外还通过不同环境中的检测实验,验证了该算法的可行性及有效性。

Obstacle detection is the main components of cross-country intelligent vehicle guidance. Cross-country environments always have changeful illuminations and complicated terrains. The paper presents a new cross-country obstacle detection method based on binocular vision system. First, we calibrated the parameters of the vision system and studied the coordination transform at first to eliminate the influence of terrains. Second the original images were preprocessed by Gaussian filter and contrast-limited adaptive histogram equalization（CLAHE） method to weaken the effect of noise, light and contrast. Harris corners were located with sub-pixel accurate. Third to guarantee the overall system real-time performance, feature-based matching techniques were studied and fundamental matrix was calculated based on random sample consensus（RANSAC）. Fourth continuity restrain was studied to eliminate pseudo matching pairs. Finally data interpolation was introduced to build elevation maps. Edge extraction and morphological processing were concerned to accomplish obstacle detection. Experimental results for different conditions are presented in support of the obstacle detection technology.