Application On The 3D Reconstruction With The Uncalibrated Image Sequences Based On The PMVS

The 3D reconstruction converts the 2D space to a 3D space,which can strongly enhance the people’s cognition to the scene.The PMVS based on the projection matrix obtained from the sparse point clouds reconstruction output a dense 3D model with the orientation information by the three steps,matching,expansion and filtering.Because the time and memory consumption are both very large, the algorithm can’t get a good result with a large dataset. Which limits the application of the algorithm.Aiming to the problem,the clusters divide algorithm was designed to enhance the processing efficiency. The redundant images have been removed and a number of compact and good size cluster of images have also been got. And then take with the PMVS on the each cluster we received. At last,a whole model is obtained by fusing the results of each cluster we got.After that,the time and the memory consumption of the whole system reduce remarkably.And a large scale dataset reconstruction can be done because of the work we have done.This thesis presents a novel method based on the image cluster dividing. The input dataset was processed before the PMVS.The redundant images can be removed from the image set to slim the dataset. And the image dataset was split into numbers of clusters which with a small size and coverage. Then take these clusters into the PMVS system in parallel. And got the results by merging these clusters together through a quality filter. Because of this, a bigger scale of the 3d reconstruction could be done.This thesis uses the information of the digital pictures, which contain the camera information such as the focal length, the size of the sensor and so on. The 3d reconstruction could process the pictures which came from the Internet,which is a really appealing thing to us and has a very great realistic meaning.