Research On The Method Of3D Recognition For The Moving Target Based On Image Sequence

3D reconstruction of image sequence is one of the important research directions incomputer vision. With the3D technology more and more widespread application,3Dreconstruction for the moving target based on the sequence images has become frontierand hotspot technologies in computer graphics, image processing and machine vision.Carried out the3D reconstruction for the moving target based on image sequence inaviation, military and civil, and so has important significance of theory and application.This thesis mainly deals with the research on how to improve the accuracy andefficiency of image matching and3D reconstruction in image sequence. The primarycoverage is as follows:1. Access to the domestic and international literature about image sequence movingtarget3D reconstruction, and introduce its research status. Elaborate the basic theory ofthe3D reconstruction based on the goals of image sequences, including the coordinates,camera model, epipolar geometry, fundamental matrix etc, as well as the principle andprocess of moving target detection.2. On the basis of the current several popular feature point extraction and matchingmethods. This paper presents an image matching algorithm based on time and frequencydomain characteristics. Firstly, the value of translation and rotation angle can be roughestimated in the frequency domain of image using Fourier transform; Then, SIFTfeature points were extracted in the time domain of the image and use the value of thetranslation and rotation angle was estimated to find the feature points to be matched;Finally, normalized correlation matching method (NCC) can be used to match images.Since the introduction of the Fourier transform, it makes full use of the frequencydomain characteristics of the image. The results displayed that, after the using of thismethod, good results have been showed.3. In order to restore the three-dimensional coordinates of feature points from threeimages, this paper proposes a linear estimation algorithm. Using the singular valuedecomposition algorithm to calculate spatial point which corresponding all the matchpoints couple in the image sequence based on triangle algorithm, and it introduces thetriangle centre of gravity as constraint condition. Then, it uses Delauny triangulation toestablish a three-dimensional network for discrete three-dimensional point.Consequently, it realizes the automated reconstruction of3D object based on multiple sequences of images. Finally, the texture mapping and color rendering be used toreconstruct a highly realistic three-dimensional target.4. Design of3D reconstruction system based on the moving target of imagesequence. This paper designs the whole system work flow and the various functionalmodules of the system. Finally, developed a3D reconstruction software based on themoving target of image sequences, besides, checked the feasibility and correctness ofthis method from the results of experiments.