Research On City Architecture Modeling From Image Sequences

3D city modeling is an essential step of realizing Digitalized City.Recently,in order to reconstruct cities more accurately and veritably,several ways are introduced: The first way apply DEM data and building structure to achieve the 3D city modeling;the second way use radar data obtained by the remote sense technology and GIS digital data to generate building models;the third way integrate the aerial and satellitic images and the point cloud data scaned with ladar to achieve urban modeling;the fourth way use shape grammar as a procedual modeling technique;the fifth way is base on images or video sequences,which use computer vision technology for city modeling.Exact models can be gained using the former three methods,but there exist some common shortcomings,such as difficulty to access data,more expensive equipments,heavy workload and the high cost.Although the fourth method is a rapid way,it is a virtual way.So it is not suitable for real cities.At present,there are a lot of research focusing on image-based 3D city modeling technology.It has merit of easily data accessing,lighter workload and lower cost.For the reasons above,this paper propose an approach to model city buildings based on image sequences and utilize the conception of mutual information in information theory.The basic idea is:we draw the outline of buildings with GoogleEarth first,then use the image sequences of the buildings along the street to construct a Multi-View panorama,using which we detect the top and bottom outlines of the building automaticly.After that,we detect and repair obscured texture region by using the concept of mutual information and a variety of method in image processing,finally we extract some enriched details from the texture to generate more visually model.Compared with other methods,our approach has the following advantages:●easier data accessing●more visually texture●extracting some details of the buildings' structure.