基于室内点云地图的特征轮廓识别方法

Feature Contour Recognition Method Based on Indoor Point Cloud Map

针对用激光雷达构建室内点云地图时存在窗户等透明物体,导致激光雷达的线束透过窗户产生噪点影响地图精度等问题,提出一种适用于室内三维点云地图的墙面识别方法,并借助此方法进一步识别出窗户的位置,进而实现对窗外的噪点与集群点的去除。前者从三维点云地图中的各个点分布的密度来识别出墙面点;后者则通过墙面坐标定位到外墙及窗户位置,从而去除噪点和无效特征。实验在三个场景下的室内地图进行方法验证,结果表明,上述方法在墙面和窗户的提取中准确率均在95%以上,并可有效的去除窗外噪点。在特定场景下窗户提取的召回率可达到100%,上述方法可用于简单和较为复杂的室内环境中进行窗墙识别去噪,在特征繁多的室内也有不错的效果。

In this paper,there are transparent objects such as windows when building indoor point cloud maps with lidar,which leads to noise generated by the lidar's wiring harness through the window and affects the accuracy of the map.To solve this problem,a wall recognition method suitable for indoor three-dimensional point cloud maps is proposed.With this method,the position of the window is further identified,and the noise and cluster points outside the window are removed.The former identifies the wall points from the density of each point distribution in the threedimensional point cloud map;the latter locates the exterior wall and window position through the wall coordinates,thereby removing noise and invalid features.The experiment verifies the method in the indoor map of three scenes.The results show that the accuracy of this method in the extraction of walls and windows is above 95%,and it can effectively remove the noise outside the window.And the recall rate of window extraction can reach 100% in specific scenarios.This method can be used for window wall recognition and denoising in simple and complex indoor environments,and has good results in indoor environments with many features.