Real-time manhattan world rotation estimation in 3D

Abstract

Abstract-Drift of the rotation estimate is a well known problem in visual odometry systems as it is the main source of positioning inaccuracy. We propose three novel algorithms to estimate the full 3D rotation to the surrounding Manhattan World (MW) in as short as 20 ms using surface-normals derived from the depth channel of a RGB-D camera. Importantly, this rotation estimate acts as a structure compass which can be used to estimate the bias of an odometry system, such as an inertial measurement unit (IMU), and thus remove its angular drift. We evaluate the run-time as well as the accuracy of the proposed algorithms on groundtruth data. They achieve zerodrift rotation estimation with RMSEs below 3.4• by themselves and below 2.8• when integrated with an IMU in a standard extended Kalman filter (EKF). Additional qualitative results show the accuracy in a large scale indoor environment as well as the ability to handle fast motion. Selected segmentations of scenes from the NYU depth dataset demonstrate the robustness of the inference algorithms to clutter and hint at the usefulness of the segmentation for further processing.

Floorplan-Jigsaw: Jointly Estimating Scene Layout and Aligning Partial Scans

“…Manhattan World (MW) assumption is the predominant rule, thus Manhattan frame estimation is well researched for both RGB [16,30] and RGB-D images [6,12]. MW assumption serves as a guidance in many applications such as layout estimation [16,30,8,3,29,40], camera pose estimation [33,13] and reconstruction refinement [7,9].…”

Efficient Global Point Cloud Alignment Using Bayesian Nonparametric Mixtures

“…6 for experiments with more than 300k points). Straub et al propose two local rotational alignment algorithms [45,44] that, similarly to the proposed approach, utilize surface normal distributions modeled as vMF mixtures. Common to all local methods is the assumption of an initialization close to the true transformation and significant overlap between the two point clouds.…”

“…Our approach to point cloud alignment relies on the fact that surface normal distributions are invariant to translation [24] and easily computed [37,44], allowing us to isolate the effects of rotation. Thus we decompose the task of finding the relative transformation into first finding the rotation using only the surface normal distribution, and then obtaining the translation given the optimal rotation.…”

“…Motivated by the observation that surface normal distributions are translation invariant [24] and straightforward to compute [37,44], we develop a two-stage branch and bound (BB) [30,31] optimization algorithm for point cloud alignment. We model the surface normal distribution of * The first two authors contributed equally to this work.…”

Robust and Globally Optimal Manhattan Frame Estimation in Near Real Time

“…, we compare exhaustive BnB and the proposed BnB with the recent MF estimation approaches: MPE[28], MMF[2], ES[10], RMFE[3], and RTMF[4]. In case of MMF, we run 10 iterations, calculate the angular errors for multiple cases, and choose a minimum error to alleviate the randomness of its initialization.…”