结合车道线检测的智能车辆位姿估计方法

基于视觉的智能车辆定位问题是自动驾驶领域研究的一大热点。在某些有效近景特征不显著的场景中,由于参与计算的特征数量不足,会导致位姿估计精度下降甚至失效。为此,本文提出一种结合车道线检测的相机位姿估计方法来提高位姿估计精度。首先,设计了一套基于自适应感兴趣区域和几何结构筛选法的车道线检测算法,精确检测到了左右车道线的内、外侧线；其次,对车道线区域内的点进行帧间匹配,得到新的匹配点对,并根据V视差图拟合出地面视差方程,求解出属于车道线匹配点对的准确视差值；最后,将这些匹配点对与ORB方法提取得到的匹配点对融合,共同参与相机的位姿计算。经实验验证,本文提出的算法提高了位姿估计结果的精度,解决了某些场景中有效特征点不足导致的位姿估计失效问题,具有良好的环境适应性。

A pose estimate of intelligent vehicle algorithm combined with lane detection

Vision-based localization of intelligent vehicles is a research hot spot in the field of autonomous driving. However, in some scenes where the effective close-range features are not significant, it is easy to cause the decrease of the pose estimation accuracy due to the insufficient of features participating in the calculation. In order to solve the problem, this paper proposes a camera pose estimation method combined with lane detection algorithm. First, a set of lane detection algorithm based on adaptive region of interest selection and geometric structure filtering methods are designed to accurately detect the inside and outside lines of the left and right lanes.Then match the points in the lane area to between to frames to obtain extra matched-point-pairs. The ground disparity equation is fitted according to the V disparity map, through which the exact disparity values of the matching point pairs belonging to the lanes can be obtained. Finally, these matched-point-pairs are merged with the matching point pairs extracted by the ORB method to participate the pose calculation of camera. It is proved that the algorithm proposed in this paper improves the accuracy of pose estimation results. It does well in solving the pose estimation failure problem caused by insufficient effective feature points in some scenes with good environmental adaptability.