Research On Remote Sensing Optical Image Based Target Spatial Registration And Texture Mapping

With the development of remote sensing technology and improving applicationneeds, compared with the target2-D(2-Dimension) imformation,3-D(3-Dimension)imformation has the more important significance no matter in the civilian or themilitary field, which provides a more perfect three-dimensional description in3-Dspace. Considering that high resolution remote sensing optical image provides moreaccurate and complete data source for the target3-D imformation.acquisition,remote sensing optical stereo imaging technology has been becoming the current hotspot. Firstly,3-D reconstruction needs target spatial registration based onMulti-Angle and Multi-temporal optical images. Space registration includes imageand stereo registration. As to photorealistic target3-D modeling, texture mapping isnecessary for adding texture imformation. This article is mainly for satellite andnear space platform, buildings, Bridges, ships and oil depot as four typical remotesensing targets. There are total three research aspects: image registration, stereoregistration and texture mapping.Firstly, in the aspect of Multi-Angle remote sensing optical image registration,the traditional methods have the advantages of large calculation amount and higherror matching rate. Therefore, space coordinate information constraints basedfeature corresponding is proposed on the basis of Correlation-like methods. Thewhole procedure of registration is from coarse to fine. The registration accuracy ismainly affected by the control points distribution. A control points selection methodbased on distributed mass is proposed. The analysis of the algorithm performance isreached by the registration experiments of Multi-Angle aerial images andMulti-Angle emulational images.Secondly, in the aspect of remote sensing target reconstruction data stereoregistration, through analyzing the data sources characteristics, considering thattraditional ICP algorithm has the problems of creating redundant matching pointsand being not easy to convergence, an improved algorithm is mentioned focus on theinvalid matching point judgement and cut. The analysis of the algorithmperformance is reached by the stereo registration experiments of three typesreconstruction data.Finally, in the aspect of remote sensing reconstruction target texture mapping, aoptical images based texture mapping framework is proposed focus on four typicaltargets. Coordinate corresponding and texture image orthorectification are studiedrespectively. The analysis of the algorithm mapping results is reached by the texture mapping experiments of four typical reconstruction targets and real two typesbulidings and do the visual evaluation of the generated3-D model.