Computer Engineering and Applications ›› 2016, Vol. 52 ›› Issue (24): 6-10.

Previous Articles     Next Articles

Real-time ground point clouds extraction algorithm using extended vertices

SUN Pengpeng, MIN Haigen, XU Zhigang, ZHAO Xiangmo   

  1. College of Information Engineering, Chang’an University, Xi’an 710064, China
  • Online:2016-12-15 Published:2016-12-20

采用延伸顶点的地面点云实时提取算法

孙朋朋,闵海根,徐志刚,赵祥模   

  1. 长安大学 信息工程学院,西安 710064

Abstract: A rapid ground point clouds extraction method using 3D-LIDAR is devised in view of the ground segmentation for autonomous vehicles in outdoor environments. First, the IMU and odometer are used to correct the distortion of LIDAR points. After that, those points are mapped to the polar grids and the extended vertices are extracted according to the consistency of those points in the vertical direction. Finally, all the points on the ground are obtained depending on the elevation of the extended vertices and the smooth of ground. The Velodyne HDL-32E is used to collect data under various scenarios to test the proposed algorithm. The results show that the proposed method works efficiently and it can great avoid over-segmentation and under-segmentation of ground. The segmentation accuracy is about 98.2% and the processing time per frame can stably control around 33 ms.

Key words: autonomous vehicle, ground point clouds extraction, 3D-LIDAR, Velodyne, polar grid map

摘要: 针对室外环境下无人驾驶车辆的地面提取实时性差的问题,提出了一种利用三维激光雷达快速提取地面的方法。首先利用车载IMU和里程计对雷达点云进行校正,然后构建柱状极坐标网格地图,根据网格中点云分布的垂直连续性提取每个网格中的延伸顶点,根据延伸顶点的高度属性以及地面平滑一致性准则提取出所有的地面点。试验中使用Velodyne HDL-32E采集不同场景下的数据作为测试集,结果表明,该方法同现有的地面分割算法相比能够降低车辆自身运动造成的提取误差,避免出现过分割和欠分割,分割准确率约为98.2%,每帧处理时间能够稳定控制在33 ms左右。

关键词: 无人驾驶车辆, 地面点云提取, 三维激光雷达, Velodyne, 极坐标网格地图