Research On Reconstruction Of Indoor Scene Based On Components Extraction

Reconstruction of 3D indoor scene is significant to automatic navigation of indoor robots,indoor virtual games and simulation experiments.The dissertation analyzes the 3D indoor scene of scanned point cloud data,extracts the components of the indoor scene,searches for the best combination from components,and completes the model matching to achieve the reconstruction of the indoor scene.The main works of this dissertation listed as follows:(1)A components extraction method based on shape decomposition from indoor scene is presented.Since the components in the indoor scene have significant shape features,the dissertation presents shape decomposition method based “slice + combination” strategy on the basis of shape extraction.Detect and Extract planes,liners and cylinders firstly,then slice the shapes by the best slicing direction,furthermore segment each slice to form visual consistent shape slices,finally using the corresponding combination strategy to combine the shape slices to achieve the extraction of the basic shape components(planar components,cylindrical components and linear components)in the door scene.Fitting the smooth curved planar components using a boundary detection-based method to maintain the boundary feature.(2)A fitting method for basic shape components of indoor scene is given.Firstly,classify the planar components according to the curvature of boundary points.Fitting the planar components with clear edges and corners using method OBB-box-based method to compensates for the lack of original data.Finally,acquire the linear and cylindrical parameters and use graphic transformation methods to fit linear and cylindrical components.(3)A recognition method based on scoring mechanism is presented.Firstly,the largest component in the basic shape components is selected as the reference component,then use the reference as the center to search for the constitute components,obtain a plurality of component combinations and establish topological structure for each component combinations.Through matching one by one between the component combination and the standard model library,and using the optimized scoring mechanism to find the component combination with the highest matching degree as the best component combination,identify the object composed by the best component combination,and use standard model replace the best component combination,and the interior scene reconstruction is finally completed.