Divide and Conquer: Efficient Density-Based Tracking of 3D Sensors in Manhattan Worlds

Zhou, Yi; Kneip, Laurent; Rodríguez, Cristian; Li, Hongdong

doi:10.1007/978-3-319-54193-8_1

Cited by 29 publications

(39 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Vanishing point [16,21] and planar structure [18,19,32,44] are two kinds of frequently-used visual cues. [18,32,44] decouple the rotation and translation to estimate orientation by tracking Manhattan frames. [19] extends to compute translational motion in VO system by minimizing de-rotated reprojection error given the rotation.…”

Section: Visual Odometry Based On Decoupled Pose Estimationmentioning

confidence: 99%

See 1 more Smart Citation

Guided Feature Selection for Deep Visual Odometry

Xue

Wang

et al. 2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

We present a novel end-to-end visual odometry architecture with guided feature selection based on deep convolutional recurrent neural networks. Different from current monocular visual odometry methods, our approach is established on the intuition that features contribute discriminately to different motion patterns. Specifically, we propose a dual-branch recurrent network to learn the rotation and translation separately by leveraging current Convolutional Neural Network (CNN) for feature representation and Recurrent Neural Network (RNN) for image sequence reasoning. To enhance the ability of feature selection, we further introduce an effective context-aware guidance mechanism to force each branch to distill related information for specific motion pattern explicitly. Experiments demonstrate that on the prevalent KITTI and ICL NUIM benchmarks, our method outperforms current state-of-theart model-and learning-based methods for both decoupled and joint camera pose recovery.

show abstract

Section: Visual Odometry Based On Decoupled Pose Estimationmentioning

confidence: 99%

“…Images in this dataset meet the Manhattan World assumption. The dataset is widely used for VO/SLAM [18,19,44] and 3D reconstruction [5]. ICL NUIM dataset is synthesized by a full 6DoF handheld camera and thus is challenging for monocular VO methods due to complicated motion patterns.…”

Section: Datasetmentioning

confidence: 99%

Guided Feature Selection for Deep Visual Odometry

Xue

Wang

et al. 2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

“…From MW surface normal vectors, [25] estimates rotational motion based on the maximum a posteriori (MAP) inference of the local Manhattan frame in real-time on a GPU. [30] decouples rotation and translation to estimate absolute orientation by tracking the Manhattan frame (MF) with a mean shift algorithm. However, this method suffers from a translation error that increases rapidly over time, as the translational motion is computed by aligning 1D density distribution of the point cloud.…”

Section: Related Workmentioning

confidence: 99%

“…where λ is a weighting factor of how certain the observation of a direction is [30]. The above procedure (lines 2 to 7 of Algorithm 1) is repeated until the change in the estimated rotation of MF is very small.…”

Section: Tracking Manhattan Framementioning

confidence: 99%

Visual Odometry with Drift-Free Rotation Estimation Using Indoor Scene Regularities

Kim

Coltin

Kim³

2017

Procedings of the British Machine Vision Conference 2017

View full text Add to dashboard Cite

We propose a hybrid visual odometry algorithm to achieve accurate and low-drift state estimation by separately estimating the rotational and translational camera motion. Previous methods usually estimate the six degrees of freedom camera motion jointly without distinction between rotational and translational motion. However, inaccuracy in the rotation estimate is a main source of drift in visual odometry. We design a hybrid visual odometry algorithm which separately estimates the rotational and translational motion to achieve improved accuracy and low drift error. To improve the accuracy of rotational motion estimation, we exploit orthogonal planar structures, such as walls, floors, and ceilings, common in man-made environments. We track orthogonal frames with an efficient SO(3)-constrained mean-shift algorithm, resulting in drift-free rotation estimates. Based on the absolute camera orientation, we newly propose a way to compute the translational motion by minimizing the de-rotated reprojection error with the tracked features. We compare the proposed algorithm with other state-of-the-art visual odometry methods and demonstrate an improved performance and lower drift error.

show abstract

“…Besides that, unlike [7,8], we also derive a closed form for the translation and analyze the limitations and what is the expected performance of the approach in a set of scene configurations. Some other interesting works assume further hypothesis in the scene geometry, as the Manhattan World assumption in [14] for scene reconstruction and in [15] for depth registration using principal component analysis of the normal vectors.…”

Section: Main Related Workmentioning

confidence: 99%

An efficient rotation and translation decoupled initialization from large field of view depth images

Martins

Fernández-Moral

Rives

2017

2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

View full text Add to dashboard Cite

Abstract-Image and point cloud registration methods compute the relative pose between two images. Commonly used registration algorithms are iterative and rely on the assumption that the motion between the images is small. In this work, we propose a fast pose estimation technique to compute a rough estimate of large motions between depth images, which can be used as initialization to dense registration methods. The main idea is to explore the properties given by planar surfaces with co-visibility and their normals from two distinct viewpoints. We present, in two decoupled stages, the rotation and then the translation estimation, both based on the normal vectors orientation and on the depth. These two stages are efficiently computed by using low resolution depth images and without any feature extraction/matching. We also analyze the limitations and observabilty of this approach, and its relationship to ICP point-to-plane. Notably, if the rotation is observable, at least five degrees of freedom can be estimated in the worst case. To demonstrate the effectiveness of the method, we evaluate the initialization technique in a set of challenging scenarios, comprising simulated spherical images from the Sponza Atrium model benchmark and real spherical indoor sequences.

show abstract

Divide and Conquer: Efficient Density-Based Tracking of 3D Sensors in Manhattan Worlds

Cited by 29 publications

References 30 publications

Guided Feature Selection for Deep Visual Odometry

Guided Feature Selection for Deep Visual Odometry

Visual Odometry with Drift-Free Rotation Estimation Using Indoor Scene Regularities

An efficient rotation and translation decoupled initialization from large field of view depth images

Contact Info

Product

Resources

About