Door And Window Detection Of Indoor Scene Based On "Global-Local" Strategy

With the popularization of 3D data acquisition devices in recent years,indoor scene reconstruction and navigation have received more and more attention.Doors and windows detection is one of the important steps,which is widely used in building protection,urban landscape,robot applications and so on.In order to detect the information of doors and windows in indoor scenes,this paper proposes a "3D-2D-3D" algorithm to automatically detect features of doors and windows in 3D indoor scenes,which adopts the "global-local" strategy to transform the detection range of doors and windows from the global area of indoor scenes to the local area containing doors and windows.Firstly,the complex 3D indoor scene is localized.A multi-view rotation photograph is taken at the center of the 3D indoor environment to generate a series of corresponding 2D images.Next,the rough detection of doors and windows in images is carried out to obtain the approximate range of doors and windows in the image.The 2D information are then mapped back to the 3D point cloud to obtain the local point cloud data including doors and windows.Finally,by processing and optimizing the contour lines and crossing points,the position information of doors and windows are obtained.The experimental results show that this "global-local" approach is efficient when detecting and identifying the location of doors and windows in 3D point cloud environment.