矿井环境基于稀疏激光雷达数据的动态物体检测与追踪

Dynamic Object Detection and Tracking Based on Sparse Lidar Data in Mine Environment

提出一种在矿井封闭昏暗环境下使用稀疏激光雷达数据的动态障碍物检测与追踪方式,利用数据特性弥补目标数据稀疏的缺点,实现检测、追踪。首先,利用静态栅格地图过滤静态数据,并使用dbscan聚类算法实现数据分类,然后,使用Graham扫描算法获得分类点云的凸包,利用凸包的角度信息对障碍物模糊分类为椭圆、矩形、直线,再利用障碍物特点选择代表障碍物的位置点;接下来,使用改进的MHT(多假设跟踪)算法完成关联数据;最终,使用多项式拟合轨迹,并预测障碍物轨迹;预测轨迹辅助移动机器人导航。实验表明,算法在封闭昏暗环境下利用稀疏数据对动态障碍物的检测与追踪,并且因为数据量少,处理迅速,适用于全方位移动机器人的导航。

In the closed dark environment of mine,a dynamic obstacle detection and tracking method using sparse lidar data is proposed,which makes up for the shortage of sparse target data by using data characteristics,and realizes detection and tracking.First,static data is filtered by static grid map,and data classification is realized by DBSCAN clustering algorithm.Then,the convex hull of classification point cloud is obtained by Graham scanning algorithm.The angle information of convex hull is used to fuzzy classify obstacles into ellipse,rectangle and straight line,and then the position points representing obstacles are selected by using the characteristics of obstacles.Next,the improved MHT(multi hypothesis tracking)algorithm is used to complete the correlation data.Finally,polynomial is used to fit the trajectory and predict the obstacle trajectory.Prediction trajectory assisted mobile robot navigation.