基于车载16线激光雷达的障碍物检测方法

针对目前车载16线激光雷达点云数据中障碍物检测算法准确率不高的问题,提出了一种基于自适应网格聚类的障碍物检测方法。首先,结合八叉树与随机抽样一致性算法(RANSAC)去除地面点;其次,投影点云至二维网格,依据各网格高程信息快速提取高结构物;然后,建立两级网格模型,按照粗网格聚类结果的分布信息自适应地确定子网格分辨率,对可能包含多目标的障碍物在子网格层进行准确检测;最后,结合相邻时刻障碍物的状态信息修正检测结果。在16线激光雷达城市道路环境测试集下的实验结果表明:该算法可准确检测行驶区域内障碍物目标,优化后的聚类算法较好地降低了欠分割与过分割错误率,检测准确率达91%。

Obstacle Detection Method Based on Vehicle 16-line Lidar

Aiming at the issue of low accuracy of the existing in obstacle detection algorithm in the vehicle 16 line Lidar point cloud data, an obstacle detection algorithm based on adaptive grid is proposed. Firstly, octree and random sample consensus (RANSAC) algorithm is utilized to remove the ground point.Secondly, project the point cloud onto the 2D-grid, tall structure objects can be quickly extracted based on the elevation information in each grid.Thirdly, a two-level grid model is established, the sub-grid resolution is determined adaptively according to the distribution information of coarse grid clustering results, the obstacles that may contain multiple targets are detected precisely at the sub-grid layer.Finally, the clustering results are improved by combining the state information of two adjacent obstacles.The experimental results under urban road environment test sets show that the proposed method can precisely detect obstacles in driving area, the optimized clustering algorithm can reduce the error rates of under-segmentation and over-segmentation,the detection accuracy is 91%.