基于图像与点云的三维障碍物检测

为提高平截头点云网络在三维障碍物检测中的精度,基于平截头点云网络的结构提出一种扩张平截头点云的检测方法。采用图像和点云数据,使用二维目标检测网络Yolov3,检测障碍物的二维包围框;扩张包围框的大小,在点云数据中提取出障碍物对应的点云;通过改进的Pointnet网络对该点云计算,得到障碍物的三维信息。在原模型基础上,加入扩张包围框,提高点云数据提取的完整性。通过KITTI数据集的验证和测试,实验结果表明,通过扩张二维包围框可以有效提高检测网络的性能。

3D obstacle detection based on image and point cloud

To improve the accuracy of the Frustum PointNets network in the detection of three-dimensional obstacles,a detection method of the expansion Frustum PointNets was proposed based on the structure of the Frustum PointNets.Image and point cloud data were used in this method and the two-dimensional target detection network Yolov3 was used to detect the two-dimensional bounding box of the obstacle.The size of the bounding box was expanded to extract the corresponding point cloud in the point cloud data.The point cloud was calculated using the improved PointNet to get the information of three-dimensional obs-tacle.Based on the original model,the expansion bounding box was added,which effectively improved the integrity of extraction from point cloud data.Through the KITTI dataset verification and testing,the experimental results show that the performance of the detection network can be effectively improved by expanding the two-dimensional bounding box.