The Least-square Matching Of Multi-view Images Based On The Photographical Feature Points

With the actual needs of targeting aerial images engineering, computer vision and digitalphotogrammetry principles, I discuss it in connection with international situation anddevelopment trend. By such characteristics as large scale production，super intelligence and highautomation, Aerial and Spatial Remote Sensing,and photogrammetry are all stepping into newera. With such background, images matching, a basic geographical information product has to beproduced precisely, automatically and rapidly. What’s more, in order to get high precisionmatching result, this paper dives deeply and systematically into techniques as obtaining andpreprocessing of matching images, multi-view matching models and strategies and Least-squaresimage matching model under different constraints. Finally I have reached a result. According tothe theory of comprehensive UCD image and the experimental results and applied research, thework accomplishments achieved and completed in this paper can be express as follows.（1）The paper introduces digital navigation camera UCD data processing of thedigital model and extracts three characteristic point operators detailed. The will provideinitial data for the matching of the experiment.The strategy and principle of six kinds ofcommon matches and application have been put forward, including the image matchingstrategies of correlation coefficient, pyramid, mountain climbing method, relevantwindow transformation and so on, and select four kinds of them to match strategy forexperiments. Through the analysis of experimental results: choose the right imagematching strategy can effectively improve the success rate, and save a lot of time.（2）I compare the two kinds of multi-view matching models，GC3and MVLL， bydifferences and similarities between the advantages and disadvantages. And the resultshows that: depending on the multi-view matching models can be integrated moreimages information, reduced the information to avoid blind mismatch similarcharacteristics. At the same time it can effectively reduce a series of problems, whichbring about repeated texture and sheltered.（3）Firstly, use available initial value coordinates depending on the least squaresmatch, then according. Then according to the different matching strategy selection andcombination, I use the correlation coefficient method for the course matching thelocalization. What’s more, adopting the least-square matching algorithm make the matching precision son-pixel precision. Through the experimental, I result verify the theadvantages and disadvantages of least-squares match.