On Bundle Adjustment for Multiview Point Cloud Registration

Multiview registration is used to estimate Rigid Body Transformations (RBTs) from multiple frames and reconstruct a scene with corresponding scans. Despite the success of pairwise pose synchronization, concept Bundle Adjustment (BA) has been proven better maintain global consistency. So in this work, we make multiview point-cloud more tractable different perspective resolving range-based BA. We first analyse optimal condition objective function BA that unifies some previous approaches. Based on analysis, propose an takes both measurement noises computational cost into account. For feature parameter update, instead calculating distribution parameters raw measurements, aggregate local distributions frame-wise fashion at each iteration. The update then only dependent number Finally, develop system using voxel-based quantization can be applied real-world scenarios. experimental results demonstrate our superiority over baselines terms accuracy speed. Moreover, also show average positioning errors achieve centimeter level. Related materials are available project page https://hyhuang1995.github.io/bareg/.