Stereo Plane R-CNN: Accurate Scene Geometry Reconstruction Using Planar Segments and Camera-Agnostic Representation

Wietrzykowski, Jan; Belter, Dominik

doi:10.1109/lra.2022.3150841

Cited by 4 publications

(6 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This work builds on results from our previous articles. A planar segment detection DNN that enables accurate geometry retrieval was introduced in [5]. We use this network in the PlaneLoc2, but add a segment geometry extraction mechanism that can be used in global localization.…”

Section: Index Terms Simultaneous Localization and Mapping Artificial...mentioning

confidence: 99%

“…Drawing from development in the object detection field, where DNNs achieve the best results, outperforming classical methods by a large margin, we also use DNN to detect reference objects in the form of planar segments. The DNN, introduced in our recent work [5] (see Fig. 3), simultaneously produces image masks of individual planar segments, their appearance descriptors used to preliminarily match segments between the local map and the global map, and retrieves the 3-D geometry of the segments.…”

Section: Planar Segments Extractionmentioning

confidence: 99%

“…It was trained on a photorealistic synthetic SceneNet Stereo dataset containing approximately 35k images from 200 different scenes. The training was started from weights pretrained on the realworld Coco and ScanNet datasets, same as in [5]. We trained the network for 10 epochs using Adam optimizer with a learning rate equal to 10 −5 and weight decay equal to 10 −4 .…”

Section: Planar Segments Extractionmentioning

confidence: 99%

“…In the PlaneLoc2 a DNN is used to segment image into planar segments and to estimate segments' supporting planes. The Stereo Plane R-CNN architecture detailed in [5] uses cameraagnostic geometry representation to provide robustness to camera parameters change and to enhance the results. To use this network for localization purposes, an export mechanism had to be added that handles the depth uncertainty.…”

Section: A Detectionmentioning

confidence: 99%

“…Correspondences between observations and instances of planar segments that serve as the target annotations are computed using 3-D mesh models, eliminating the need for tedious manual labeling. Such a modification adds little overhead to the Stereo Plane R-CNN model from [5], while producing discriminative descriptors.…”

Section: B Descriptionmentioning

confidence: 99%

See 4 more Smart Citations

PlaneLoc2: Indoor Global Localization Using Planar Segments and Passive Stereo Camera

Wietrzykowski

2022

IEEE Access

Self Cite

View full text Add to dashboard Cite

This paper introduces PlaneLoc2 -a novel indoor global localization system designed to harness the potential of stereo cameras. A need for robust global localization that does not produce incorrect results (false positives) is present in almost every life-long autonomy task. We show that planar segments extracted from stereo vision data by a neural network enable such robust localization. Planar segments are easier to discriminate than keypoint features and provide easy-to-use geometric constraints. We propose an architecture that exploits a single deep neural network (DNN) to detect planar segments, produce appearance descriptors, and estimate segment geometry. Moreover, we introduce a novel view-based segment map and a novel pose retrieval procedure that considers the uncertainty of features to efficiently use the geometric constraints provided by them. We also show that the new learned descriptor provides better discrimination than the hand-crafted one. Finally, we present experimental results that show that our solution outperforms other state-of-the-art global localization methods and does not produce incorrect agent poses. For both test scenes it recognizes at least 15% more poses than the second best method without incorrect recognitions.

show abstract

Section: Index Terms Simultaneous Localization and Mapping Artificial...mentioning

confidence: 99%

Section: Planar Segments Extractionmentioning

confidence: 99%

Section: Planar Segments Extractionmentioning

confidence: 99%

Section: A Detectionmentioning

confidence: 99%

Section: B Descriptionmentioning

confidence: 99%

See 3 more Smart Citations

PlaneLoc2: Indoor Global Localization Using Planar Segments and Passive Stereo Camera

Wietrzykowski

2022

IEEE Access

Self Cite

View full text Add to dashboard Cite

show abstract

VoxelPlane-Reloc: an indoor scene voxel plane relocalization algorithm

Suo,

Wang,

Huang

et al. 2024

Complex Intell. Syst.

View full text Add to dashboard Cite

We present VoxelPlane-Reloc, a novel indoor plane relocalization system based on voxel point clouds, designed for use with depth cameras. First, we propose an adaptive weighted plane extraction model that allows for dynamic adjustment of the correlation between points and plane accuracy. Second, we construct a plane merging model based on voxel growth, which employs a voxel neighborhood growth strategy to handle unmerged planes and allows for the merging of under-growing planes. Third, we present an incremental approach for plane input and propose a strategy for triplet selection and evaluation based on the structural constraints of the planes. This system relies solely on point clouds for relocalization and does not depend on other information, such as RGB data. We extensively evaluate the system on four datasets, and the experimental results demonstrate that the system can accurately and quickly perform relocalization with an average precision of $$99.37\%$$ 99.37 % . The time for relocalization is improved by $$92.43\%$$ 92.43 % compared to previous plane relocalization systems, and it exhibits strong robustness to indoor plane structures.

show abstract

A review of visual SLAM with dynamic objects

Qin,

2023

View full text Add to dashboard Cite

Purpose This paper aims to provide a better understanding of the challenges and potential solutions in Visual Simultaneous Localization and Mapping (SLAM), laying the foundation for its applications in autonomous navigation, intelligent driving and other related domains. Design/methodology/approach In analyzing the latest research, the review presents representative achievements, including methods to enhance efficiency, robustness and accuracy. Additionally, the review provides insights into the future development direction of Visual SLAM, emphasizing the importance of improving system robustness when dealing with dynamic environments. The research methodology of this review involves a literature review and data set analysis, enabling a comprehensive understanding of the current status and prospects in the field of Visual SLAM. Findings This review aims to comprehensively evaluate the latest advances and challenges in the field of Visual SLAM. By collecting and analyzing relevant research papers and classic data sets, it reveals the current issues faced by Visual SLAM in complex environments and proposes potential solutions. The review begins by introducing the fundamental principles and application areas of Visual SLAM, followed by an in-depth discussion of the challenges encountered when dealing with dynamic objects and complex environments. To enhance the performance of SLAM algorithms, researchers have made progress by integrating different sensor modalities, improving feature extraction and incorporating deep learning techniques, driving advancements in the field. Originality/value To the best of the authors’ knowledge, the originality of this review lies in its in-depth analysis of current research hotspots and predictions for future development, providing valuable references for researchers in this field.

show abstract

Stereo Plane R-CNN: Accurate Scene Geometry Reconstruction Using Planar Segments and Camera-Agnostic Representation

Cited by 4 publications

References 24 publications

PlaneLoc2: Indoor Global Localization Using Planar Segments and Passive Stereo Camera

PlaneLoc2: Indoor Global Localization Using Planar Segments and Passive Stereo Camera

VoxelPlane-Reloc: an indoor scene voxel plane relocalization algorithm

A review of visual SLAM with dynamic objects

Contact Info

Product

Resources

About