应用激光雷达与相机信息融合的障碍物识别

Obstacle recognition based on lidar and camera information fusion

摘要:

针对探测器在地外星体表面软着陆过程中的障碍物识别问题,提出了一种融合三维点云数据与灰度图像数据进行精确障碍物识别的方法。首先利用坐标转换将灰度图像与三维点云归一化到同一坐标系下,实现传感器数据的融合。然后采用改进K均值聚类算法对预处理后灰度图像进行图像分割,生成光学障碍图。最后利用开源库PCL(Point Cloud Library)对激光雷达生成的三维激光点云数据进行处理,采用随机采样一致性算法提取着陆区地形水平面,对去除水平面后的点云数据进行点云分割,分离出突起物、凹坑等障碍物,并通过激光雷达与相机转换坐标系,投影到像平面,生成最终障碍图。

Abstract:

Aiming at the obstacle recognition problem of the detector in the soft landing process of the extraterrestrial surface, a method of fusing 3D point cloud and grayscale image for accurate obstacle recognition is proposed. Firstly, the grayscale image and the 3D point cloud are normalized to the same coordinate system by coordinate transformation. Then the K-means clustering algorithm is used to the pre-processed grayscale image to generate the optical obstacle map. Finally, the 3D point cloud which generated by the lidar is processed by the open source library PCL (Point Cloud Library).The RASAC (Random Sample Consensus, RASAC) algorithm is adopted to extracts the plane of the landing zone. After removing the plane from the point cloud, a segmentation algorithm is used to separate obstacles such as protrusions and pits. By the convert matrix from lidar coordinate to camera coordinate, the obstacles are projected to the grayscale image which the final obstacle map is generated.