Patchwork++: Fast and Robust Ground Segmentation Solving Partial Under-Segmentation Using 3D Point Cloud

Lee, Seungjae; Lim, Hyungtae; Myung, Hyun

doi:10.1109/iros47612.2022.9981561

Cited by 71 publications

(38 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Nonetheless, we highlight that BEV is often sufficient for various autonomous driving tasks such as obstacle avoidance, motion forecasting, and path planning. We believe that further improvement in 3D box localization can be achieved through leveling the pseudolabel boxes with ground plane segmentation [59], [60] of all accumulated sequence frames, or using a 3D map [61]. Additionally, we notice that training a single multi-frame detector on its own pseudo-labels (ST3D) can lead to a degradation in performance compared to the Source-only approach for all three target domains.…”

Section: Resultsmentioning

confidence: 97%

See Eye to Eye: A Lidar-Agnostic 3D Detection Framework for Unsupervised Multi-Target Domain Adaptation

Tsai

Berrío

Shan

et al. 2022

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

We introduce Multi-Source 3D (MS3D), a new self-training pipeline for unsupervised domain adaptation in 3D object detection. Despite the remarkable accuracy of 3D detectors, they often overfit to specific domain biases, leading to suboptimal performance in various sensor setups and environments. Existing methods typically focus on adapting a single detector to the target domain, overlooking the fact that different detectors possess distinct expertise on different unseen domains. MS3D leverages this by combining different pre-trained detectors from multiple source domains and incorporating temporal information to produce high-quality pseudolabels for fine-tuning. Our proposed Kernel-Density Estimation (KDE) Box Fusion method fuses box proposals from multiple domains to obtain pseudo-labels that surpass the performance of the best source domain detectors. MS3D exhibits greater robustness to domain shifts and produces accurate pseudolabels over greater distances, making it well-suited for highto-low beam domain adaptation and vice versa. Our method achieved state-of-the-art performance on all evaluated datasets, and we demonstrate that the choice of pre-trained source detectors has minimal impact on the self-training result, making MS3D suitable for real-world applications. Our code is available at https://github.com/darrenjkt/MS3D.

show abstract

Section: Resultsmentioning

confidence: 97%

See Eye to Eye: A Lidar-Agnostic 3D Detection Framework for Unsupervised Multi-Target Domain Adaptation

Tsai

Berrío

Shan

et al. 2022

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

show abstract

“…The results indicate that our methods outperform the state-of-theart (SOTA) works (Shan et al, 2020;Xu et al, 2022;Koide et al, 2019;Lee et al, 2022). The contributions of this work can be summarized as follows:…”

Section: Introductionmentioning

confidence: 88%

SLAM-RAMU: 3D LiDAR-IMU lifelong SLAM with relocalization and autonomous map updating for accurate and reliable navigation

Chen,

Zhong,

Xie

et al. 2024

View full text Add to dashboard Cite

Purpose Autonomous mobile robots (AMRs) play a crucial role in industrial and service fields. The paper aims to build a LiDAR-based simultaneous localization and mapping (SLAM) system used by AMRs to overcome challenges in dynamic and changing environments. Design/methodology/approach This research introduces SLAM-RAMU, a lifelong SLAM system that addresses these challenges by providing precise and consistent relocalization and autonomous map updating (RAMU). During the mapping process, local odometry is obtained using iterative error state Kalman filtering, while back-end loop detection and global pose graph optimization are used for accurate trajectory correction. In addition, a fast point cloud segmentation module is incorporated to robustly distinguish between floor, walls and roof in the environment. The segmented point clouds are then used to generate a 2.5D grid map, with particular emphasis on floor detection to filter the prior map and eliminate dynamic artifacts. In the positioning process, an initial pose alignment method is designed, which combines 2D branch-and-bound search with 3D iterative closest point registration. This method ensures high accuracy even in scenes with similar characteristics. Subsequently, scan-to-map registration is performed using the segmented point cloud on the prior map. The system also includes a map updating module that takes into account historical point cloud segmentation results. It selectively incorporates or excludes new point cloud data to ensure consistent reflection of the real environment in the map. Findings The performance of the SLAM-RAMU system was evaluated in real-world environments and compared against state-of-the-art (SOTA) methods. The results demonstrate that SLAM-RAMU achieves higher mapping quality and relocalization accuracy and exhibits robustness against dynamic obstacles and environmental changes. Originality/value Compared to other SOTA methods in simulation and real environments, SLAM-RAMU showed higher mapping quality, faster initial aligning speed and higher repeated localization accuracy.

show abstract

“…By removing the ground, we find better correspondence between local features in global place recognition. There are several methods to remove the ground, including RANSAC-based plane estimation, deep learning-based methods, and using a probability model to estimate the ground [29], [30], [31].…”

Section: A Ground Preprocess For Global Place Recognitionmentioning

confidence: 99%

“…We use [5] as the baseline for evaluating the impact on the performance of the proposed place recognition method. For preprocessed ground training, we remove the ground that uses the probability model [31] to estimate the ground and only use non-ground points for training. In training using L CAU X in Eqn.…”

Section: Experiments a Implementation And Setupmentioning

confidence: 99%

HI-FFT: Heterogeneous Parallel In-Place Algorithm for Large-Scale 2D-FFT

Kang

Lee

Kim³

2021

IEEE Access

View full text Add to dashboard Cite

Fast Fourier Transform (FFT) is a fundamental operation for 2D data in various applications. To accelerate large-scale 2D-FFT computation, we propose a Heterogeneous parallel In-place 2D-FFT algorithm, HI-FFT. Our novel work decomposition method makes it possible to run our parallel algorithm on the original data (i.e., in-place), unlike prior parallel algorithms that require additional memory space (i.e., out-of-place) to guarantee independence among sub-tasks. Our work decomposition method also removes the duplicated operations on the out-of-place approaches. Using our decomposition method, we introduced an in-place heterogeneous parallel algorithm that utilizes both multi-core CPU and GPU simultaneously. To maximize the utilization efficiency of the computing resources, we also propose a priority-based dynamic scheduling method. We compared the performance of seven different 2D-FFT algorithms, including ours, for large-scale 2D-FFT problems whose sizes varied from 20K 2 to 120K 2 . As a result, we found that our method achieved up to 2.92 and 4.42 times higher performance than the conventional homogeneous parallel algorithms based on the state-of-the-art CPU and GPU libraries, respectively. Also, our method showed up to 2.27 times higher performance than the prior heterogeneous algorithms while requiring two times less memory space.To check the benefit of our HI-FFT on an actual application, we applied it to a CGH (Computer Generated Holography) process. We found that it successfully reduces the hologram generation time. These results demonstrate the advantage of our approach for large-scale 2D-FFT computation.

show abstract

Patchwork++: Fast and Robust Ground Segmentation Solving Partial Under-Segmentation Using 3D Point Cloud

Cited by 71 publications

References 25 publications

See Eye to Eye: A Lidar-Agnostic 3D Detection Framework for Unsupervised Multi-Target Domain Adaptation

See Eye to Eye: A Lidar-Agnostic 3D Detection Framework for Unsupervised Multi-Target Domain Adaptation

SLAM-RAMU: 3D LiDAR-IMU lifelong SLAM with relocalization and autonomous map updating for accurate and reliable navigation

HI-FFT: Heterogeneous Parallel In-Place Algorithm for Large-Scale 2D-FFT

Contact Info

Product

Resources

About