3D Reconstruction Based On Multiple No-order Images

3D reconstruction is one of the most popular research field in the realm of computer vision,which restores object’s three-dimensional information in the space through its redundant two-dimensional information. Its inputs can be multiple no-order images which are taken from an single camera to multiple directions of the reconstructed object, while it can also be from different cameras. Projective geometry and camera imaging principle is the theoretical basis. We also introduce homography matrix, essential matrix and fundamental matrix’s concepts, computational methods and special application, while the computational methods are related to four-point algorithm, eight-point algorithm, directed linear transformation, ransom sample consensus, the least square method and so on. We analyse the classic sift features extraction algorithm and the corresponding matching method. The themes of this article are discussed as follows:On the aspect of matching features, we compare the traditional linear scanning algorithm and BBF algorithm based on k-dimensional tree. On the aspect of deleting mismatched pairs, we simplify the traditional process, which obtains a middle model utilizing RANSAC and applies it to the original set of pairs while the outer points are just the deleting pairs and applies the least square method to the inner points to get the final model. Our algorithm keeps the inner points as matching pairs directly from RANSAC and applies the least square method to them to obtain the final model, which promotes time efficiency in the view of not impacting the number of matching pairs and mismatching errors.We go deep into the algorithm of relating the features of multiple images, which utilizes pair and vector, and cooperates with BFS,tags and defensive programming to relate multiple no-order images.This algorithm is the one that is indispensable in the process of 3D reconstruction based on multiple no-order images.The perspective of camera imaging principle is helpful for the process of 3D reconstruction, especially on the restoration of translation vector from camera matrix, we introduce a new method. We start from the formula)36.2(, the left side 33 ? matrix of the camera matrix is a reversible matrix M, and the right side matrix is ??tM, we can get the translation matrix from inverse and multiply. This method is simple and clear while the implementation is easy and less error-prone.3D reconstruction based on multiple no-order images. We go deep into this gradually increasing 3D reconstruction process,which includes the choice of the new increased camera, its restoration of intrinsic and extrinsic parameters, and the extension of sparse point-cloud, and error control and management. In the end, we show the result of the restored sparse point-cloud.