The Research Of Indoor 3D Scene Modeling Based On Multi-view Images

The data procurement and 3D modeling of the indoor scene is so important for building of the Smart City.3D modeling methods based on the close-range image sequence use the normal digital camera as the image acquisition device,which has lower cost,more efficient,lower labor intensity,and could easier to adapt to the indoor scene.However,due to the images of indoor scene have fateful shading and severe perspective distortion,with some problems,like single texture and more repeat,are contained by the indoor scene,providing more difficulties of extracting and matching of the image information.Thus,it is urgent to improve the modeling and precision of the indoor scene.In the image information,the point feature has precise location,which could recover the mapping relation between the 2D image and 3D geometry,effectively;the line feature is the main outline feature of the indoor model,having the control effect for the whole indoor model.Because of the multivariate indoor scene,it is hard to satisfy a variety of indoor conditions when using the single modeling method.Therefore,this paper,based on the image sequence modeling method,has researched the feature points matching,extracting,and the modeling of 3D outline model under different situations.The principal results and conclusions as follows:(1)According to different indoor situations,if indoor images have more textures and features are more visible,the SIFT operator would be used to extract and match indoor image feature points.However,if some cases,like the roughcast house,do not have feature points to some extent,this paper uses the artificial adding feature point method to assisted extract and match feature points.(2)On the basis of extracting more real and credible feature points,calculating locations of 3D points,according to the characteristic of the close-range image feature point,based on the optimization principle of beam adjustment synchronously calculated by camera parameters and 3D point locations.In addition,this paper combines the space resection of single image method,if the imagery sequence could not provide enough feature points to auxiliary calculate camera parameters in doors.This paper calculates camera parameters of the single image directly.Aimed at indoor different situations,using different methods could calculate camera parameters of indoor images effectively.(3)Given that the importance of line features to indoor 3D scene models,it is the key to the indoor 3D scene model whether to establish an accurate indoor outline model.This paper based on the line feature extraction uses the multi-constrained line feature matching method,applying the lines obtained from matching to the 3D rebuilding of line features based on the principle of ensuring spatial lines by surface-surface intersection,rebuilding the line feature.On the basis of this research,the 3D outline model using this method is hard to recover the indoor 3D outline model sometimes,thus this paper also propose the way of using man-machine interaction for artificial extracting and matching to the indoor outline.This way could model the indoor outline effectively.If acquiring the indoor outline,this paper could control the whole indoor scene effectively.This is the basis of build the whole indoor 3D scene model.(4)After building the indoor linear frame,this paper combines many existing modeling methods of the entity model,modeling indoor entities with different situations,different objects,and different methods.This paper has built some more accuracy and more complete indoor entity 3D models.The basic elements of indoor scene consist of indoor frame models and entity models.On the basis of already built indoor linear frame models,this paper adds already built entity models,respectively,and finally,constitutes the indoor 3D scene model.