基于视觉理论的动态点云剔除算法

Dynamic point cloud removal algorithm based on vision theory

针对动态场景下构建的点云地图中包含大量动态目标的错误点云问题,提出一种基于视觉理论将三维点云转换视觉图像的动态点云剔除算法。通过对当前帧和包含动态点云的噪声地图做点云的地面分割和高度分割,将点云的深度信息转换成视觉可用的图像信息,利用视觉理论中的背景差分法对当前帧和噪声地图进行深度图像对比,筛选出初始动态点云并计算动态分数;根据动态分数对初始动态点云进行自适应最近邻搜索以剔除动态目标。实验结果表明,所提算法的动态点云剔除率可达94%以上,整体得分为96.34,能有效剔除场景中的动态目标。

Aiming at the problem that the point cloud map constructed in the dynamic scene contains a large number of dynamic objects,a dynamic point cloud elimination algorithm based on vision theory to convert 3D point clouds into visual images was proposed.Through the current frame and the noise map containing dynamic point cloud,the ground and height segmentation of point cloud were made.Then the depth information of the point cloud was converted into visual image information,the background difference method in vision theory was used to compare the depth image of the current frame and the noise map,and the initial dynamic point cloud was screened out and the dynamic score was calculated.Finally,adaptive nearest neighbor search was carried out on the initial dynamic point cloud according to the dynamic score to eliminate the dynamic target.Experimental results showed that the dynamic point cloud removal rate of the proposed algorithm could reach more than 94%,and the overall score was 96.34,which could effectively eliminate dynamic targets in the scene.