The Research Of Sequence Image Projective Rconstruction

Virtual reality technology has established a deep connection with the modern industry and social life. With the development of the theory and the advance in technology, people are eager to roam the3D scenes of real world, rather than satisfied only by the virtual models generated by the computer-aided design software. Now, multiple view geometry is one of the most promising theories to solve this problem. Projective reconstruction, of which image matching plays a decisive impact on the accuracy, is an essential foundation of the applications in the multiple view geometry.Image correspondence is a key problem in computer vision. Although many matching technique in short-base-line have been developed, the wide-baseline correspondence problem with large scale, rotation, illumination and affine transformations is still not tackled very well. The paper analysis the defect of the traditional method proposed a new matching method which base on multi-scale Harris algorithm and guide matching to achieve large number of accurate point correspondences between uncalibrated image sequences of the same scene for wide baseline. Experimental results show that the guide matching method can be used for severe scene variations and provide evidence of improved performance with respect to the SIFT distance and Harris matchers. It is useful to the matching in short-base-line also, and the results of this method are better than the traditional method.At the projective reconstruction,we introduced two projective reconstruction methods:projective reconstruction based on SVD decomposition and epipolar constraint, as well as the integration of these two methods, and the second one is more important in this paper.In addition, in order to meet the needs of the experiment in the application, as little as possible of the camera is used in the projective reconstruction, so for the large baseline matching, the angle to match is increased to a very precise image match the parallactic angle of85degrees. Furthermore, CUDA parallel computing are available in the feature point detection algorithm, guide matching and projective reconstruction algorithms to accelerate.