Research On Object Extraction Method Of Indoor Scene Based On Point Cloud

Indoor scene object extraction has a wide range of applications in intelligent robots,computer vision intelligence and other fields..It has great significance for scene modeling,target detection and scene analysis.Most of the current object extraction method is based on a complete point cloud model of the scene.However,the single viewpoint of the point cloud model is consistent with the human visual model.The current methods has certain limitations on the extraction of objects in scenes expressed by such point cloud models.Therefor,the following research work was carried out based on the single viewpoint and unilateral point clout scene data as the research object.(1)A method for preprocessing the point cloud scene which are measured from single-viewpoint is implemented.Construct a KD-Tree,then remove outliers based on KD-Tree and eliminating noise interference.Use RANSAC algorithm to detect all planes in the scene,estimated the rotation direction of the scene,given the rotation transformation matrix,and then complete the Erection transformation of the scene.(2)Gives a projection-based scene segmentation method.After rotate the scene to the natural vertical direction.The plane with the lowest horizontal plane in the scene is the ground,and the plane perpendicular to the ground and the outermost plane in the scene is the wall surface.First removes the floor and walls from the scene,then project the remaining points to two-dimensional coordinate system in the ground plane.The use of distance-based region growing algorithm for the projection results for area clustering on the plane.Obtaining the segmentation of a three-dimensional point set of the scene by reverse mapping.(3)An object extraction method based on scene division is designed.For each of the divided scenes(called sub-scenes),first detect the largest horizontal plane and extract the point set in the normal direction.Segment these points using a plane detection algorithm,and judge the concavity into one object.(4)A method of scene expression and scene topology construction is given.The object is represented by using the center point of the object as a node,and the optimization of the object extraction to the bounding box is adopted.Using the geometry to replace the corresponding obj ect,and finally use the shortest path between nodes to build the scene topology.