南方基地知识库

近年来,自动驾驶技术得到了加速发展,提出许多经典的姿态估计算法实现车辆的定位.但在实际场景中,一些运动点会给姿态估计结果带来误差.提出了一种基于运动去除的姿态估计算法.首先,基于特征点信息建立了几何约束条件,实现点的初步筛选.其次,利用聚类算法将视差图划分为不同的聚类区域.随后,根据数学模型将聚类区域判断为动态区域或静态区域,进一步丢弃动态区域中的特征点.实验表明,本文算法能够有效去除错误的特征信息,更加准确的实现车辆定位.

Recently, the autonomous driving technology has been developing rapidly, proposing many classical pose estimation algorithms to realize vehicle positioning. However, some moving points in reality will bring errors to the pose estimation algorithm. With the purpose of improving positioning accuracy, this paper brings forward a pose estimation algorithm based on motion removal. At first, establish the geometric constraints to realize selection of points preliminarily based on the feature points. Secondly, divide the disparity map into different clustering regions by the clustering algorithm. Sequentially, divide the clustering region into dynamic region and static region according to the mathematical model. Meanwhile, discard the feature points in the dynamic region. The result shows that the algorithm in this paper can effectively identify and remove the wrong feature information, realizing the more accurate vehicle positioning successfully.