This paper presents a region-based method for extraction of consistent surfaces from raw point clouds. The method uses a new robust estimation method of constructing seed regions and a new method of orientating region...This paper presents a region-based method for extraction of consistent surfaces from raw point clouds. The method uses a new robust estimation method of constructing seed regions and a new method of orientating regions or surfaces. The robust estima- tion method selects good seed regions from candidate regions generated randomly in a structured neighborhood. The orienta- tion method uses transition vectors from which include angles of adjacent normal vectors are not greater than 90~ and thus can be orientated correctly crossing sharp features or close-by opposite surfaces. The region-based method consists of two levels of segmentation: planar segmentation and quadric segmentation, both of which produce consistent surfaces. The quadric segmen- tation fits general quadrics by 3 L fitting algorithm in its region growing process and can take consistent planar surfaces as ini- tials. Experimental results show that the robust estimation method has higher probability of success than the traditional one and the orientation method works well. Experimental results also demonstrate the applicability of our method to various data.展开更多
针对当前三维目标检测中存在的数据降采样难、特征提取不充分、感受野有限、候选包围盒回归质量不高等问题,基于3DSSD三维目标检测算法,提出了一种基于原始点云、单阶段、无锚框的三维目标检测算法RPV-SSD(random point voxel single st...针对当前三维目标检测中存在的数据降采样难、特征提取不充分、感受野有限、候选包围盒回归质量不高等问题,基于3DSSD三维目标检测算法,提出了一种基于原始点云、单阶段、无锚框的三维目标检测算法RPV-SSD(random point voxel single stage object detector),该算法由随机体素采样层、3D稀疏卷积层、特征聚合层、候选点生成层、区域建议网络层共五个部分组成,主要通过聚合随机体素采样的关键点逐点特征、体素稀疏卷积特征、鸟瞰图特征,进而实现对物体类别、3D包围盒以及物体朝向的预测。在KITTI数据集上的实验表明,该算法整体表现良好,不仅能够命中真值标签中的目标并且回归较好的包围盒,还能够从物体的不完整点云推测出物体的类别及其完整形状,提高目标检测性能。展开更多
基金supported by the National Natural Science Foundation of China(Grant No.51205332)the SRF for the Returned Overseas Chinese Scholars
文摘This paper presents a region-based method for extraction of consistent surfaces from raw point clouds. The method uses a new robust estimation method of constructing seed regions and a new method of orientating regions or surfaces. The robust estima- tion method selects good seed regions from candidate regions generated randomly in a structured neighborhood. The orienta- tion method uses transition vectors from which include angles of adjacent normal vectors are not greater than 90~ and thus can be orientated correctly crossing sharp features or close-by opposite surfaces. The region-based method consists of two levels of segmentation: planar segmentation and quadric segmentation, both of which produce consistent surfaces. The quadric segmen- tation fits general quadrics by 3 L fitting algorithm in its region growing process and can take consistent planar surfaces as ini- tials. Experimental results show that the robust estimation method has higher probability of success than the traditional one and the orientation method works well. Experimental results also demonstrate the applicability of our method to various data.
文摘针对当前三维目标检测中存在的数据降采样难、特征提取不充分、感受野有限、候选包围盒回归质量不高等问题,基于3DSSD三维目标检测算法,提出了一种基于原始点云、单阶段、无锚框的三维目标检测算法RPV-SSD(random point voxel single stage object detector),该算法由随机体素采样层、3D稀疏卷积层、特征聚合层、候选点生成层、区域建议网络层共五个部分组成,主要通过聚合随机体素采样的关键点逐点特征、体素稀疏卷积特征、鸟瞰图特征,进而实现对物体类别、3D包围盒以及物体朝向的预测。在KITTI数据集上的实验表明,该算法整体表现良好,不仅能够命中真值标签中的目标并且回归较好的包围盒,还能够从物体的不完整点云推测出物体的类别及其完整形状,提高目标检测性能。