Multicore bundle adjustment

Abstract

We present the design and implementation of new inexact Newton type Bundle Adjustment algorithms that exploit hardware parallelism for efficiently solving large scale 3D scene reconstruction problems. We explore the use of multicore CPU as well as multicore GPUs for this purpose. We show that overcoming the severe memory and bandwidth limitations of current generation GPUs not only leads to more space efficient algorithms, but also to surprising savings in runtime. Our CPU based system is up to ten times and our GPU based system is up to thirty times faster than the current state of the art methods [1], while maintaining comparable convergence behavior. The code and additional results are available at http://grail.cs. washington.edu/projects/mcba. Jx J c , J p 0.023 0.044 J c , J p 0.016 0.078 J T y J T c , J p 0.035 0.076 J c , J p 0.055 0.11 J p J c 0.032 0.16 J c , J p 0.047 0.52 (*) M −1 λ J T c , J p 0.037 0.077 J c , J p 0.052 NA J p J c 0.025 0.15 J c , J p 0.041 1.34 (*)

Image Based 3D Reconstruction of Texture-less Objects for VR Contents

“…The feature matching algorithm implemented in our methodology used the preemptive matching technique that accelerates the matching process and reduces the reasonable amount of time [19][20]. It arranges all the features of each image into decreasing scale order and enlists all the pairs to be matched.…”

Dense 3D Reconstruction with a Hand-Held Camera

“…The parts are then recursively merged to obtain the final reconstruction. For bundle adjustment we use the implementation of Wu et al [20]. After adding a new camera, we generate new points from the trajectories.…”

Lighting transfer across multiple views through local color transforms

“…Moreover, we can reconstruct a sparse point cloud from multi-view photos and find correspondences between images, which allows local analysis of lighting changes. We use offthe-shelf VisualSfM [25]: we first apply structure from motion [26] to estimate the parameters of the cameras and then use patch-based multi-view stereo [27] to generate a 3D point cloud of the scene. For each point, the algorithm also estimates a list of images in which it appears.…”