Research On Target3D Reconstruction Technology Based On Image Sequences

With using the digital camera as image sensor of original2D images,3Dreconstruction of the target based on image sequences completes the non-contact3Dmeasurement with image processing and computer vision technology, then obtains the3D information automatically through programming. This technology simulates themethod of human visual processing scene, it can receive the3D information indifferent kinds of situation and can’t be replaced by other3D reconstruction methods.Compared with the traditional method such as the modeling software and3D scanner,3D reconstruction based on the image sequences is lower cost, more realistic, higherdegree of automation, so this method will have a wide application prospects.The paper introduces the basic concepts and theory of the3D reconstructioncontains the basic theory of projective geometry, camera model, imaging principle,coordinate transformation, camera calibration, image preprocessing, feature extractionand matching, fundamental matrix, essential matrix and cluster adjustment. Zhang’scalibration method is selected as the calibration method of the paper after discussingand analying. The paper discusses several current filters and edge detection methodsin detail, and gives the results of some methods. In this way to removal the uselessinformation and keep the useful information and build good foundation for3Dreconstruction. The paper discusses several feature extraction and matching methodsin detail, through comparision, the feature extraction and matching method based onSURF is selected as the method of the paper. Then put forward a feasible proposalaimed the3D reconstruction based on the image sequences, and analyse every step indetail. This paper discusses how to connect the several image sequences using thefeature extraction and matching, how to reconstruct the first two images using the twoimage reconstruction algorithm and then use it as an initial structure, how to get theprojection matrix of the new added image through feature point tracking relationship,the process of the new added matching point, the process of the new added3D points,cluster adjusted and so on. Finally, the paper gives the results and3D point cloud, atthe same time, also gives the experiment error and error analysis.