Background
The estimation of camera pose and the generation from the sparse points, the basis for dense matching and texture mapping, are very important in 3D reconstruction. There are two main methods, one is the overall reconstruction, and the other is iterative reconstruction. The overall reconstruction processes all the images together and avoids the problem of local optimization, however its calculation is very large and once a new picture is added, all the images need to be reprocessed.
While iterative reconstruction uses the relationship between multiple images and perform incremental computation. When a new image is added, only this new one is required to be calculated. The disadvantage of iterative reconstruction is the cumulative error caused by the calculation of the image based on the specified sequence. In addition, much of time is consumed in the process of image matching. So when reconstructing from large-scale images, it often causes slow operation and even failure of reconstruction.
Therefore, the availability and capacity of the image storage system have become the two important issues in the calculation. The real availability and good performance of the automatic elastic telescopic storage system can not only solve the problems in storage system, like machine failure, hard disk damage and user data overflow, but also reduce the risk of workload, and ensure system be more durable and stable.
So, in this paper, we focus on how to introduce the high-performance storage and calculation in the iterative reconstruction.