检索结果-内蒙古大学图书馆

LiDAR semantic segmentation with local consistency constrained KPConv LSTM

NEUROCOMPUTING 2025年 626卷

作者： Bai, Tingming Xiang, Zhiyu Zhao, Xijun Xu, Peng Pu, Tianyu Fu, Jingyun Zhejiang Univ Coll Informat Sci & Elect Engn Hangzhou 310027 Zhejiang Peoples R China China North Artificial Intelligence & Innovat Res Beijing 100072 Peoples R China

Asa fundamental task for autonomous driving, LiDAR point cloud semantic segmentation has been intensively studied in recent years. Despite the great progress, achieving satisfactory semantic segmentation is still very challenging due to the sparsity of LiDAR points and the shape diversity of the classes in the open world. In this paper we propose a local consistency constrained KPConv-LSTM module to benefit the existing methods. It enhances point-based LSTM with several KPConvs to strengthen and align the previous hidden features, thus improving the temporal feature propagation. A temporal weighting block is designed within the module to further reduce the error caused by the misalignment of moving objects. In addition, a special local consistency loss is proposed to encourage the local smoothness of the feature, thereby providing more consistent feature for temporal propagation in LSTM. We apply our method to various existing LiDAR semantic segmentation models. The experimental results on multiple datasets show that our method can produce notable improvements on all of them, validating the effectiveness and generality of the method.

关键词： Semantic scene understanding deep learning for visual perception LiDAR semantic segmentation Point cloud processing LSTM mechanism

来源：评论

学校读者我要写书评

暂无评论

H-PCC: Point Cloud Compression With Hybrid Mode Selection and Content Adaptive Down-Sampling

引用

IEEE ROBOTICS AND AUTOMATION LETTERS 2025年第4期10卷 3054-3061页

作者： Liu, Bowen Chen, Yu Wang, Boyang Yang, Mingyu Kim, Hun-Seok Univ Michigan EECS Dept Ann Arbor MI 48109 USA

LiDAR sensors are integral to autonomous driving and augmented reality applications, providing essential depth information. However, managing the substantial volume of LiDAR point cloud data is crucial for practical application, necessitating efficient compression algorithms. Similar to other data compression domains, point cloud compression aims to spatially and temporally decorrelate information while preserving reconstruction quality. Addressing real-world challenges such as sensor pose retrieval inaccuracies or motion estimation failures, we introduce H-PCC, a hybrid LiDAR data compression pipeline that combines a range image prediction-based dynamic compression method with a novel point clustering octree-based static method. Additionally, we demonstrate that our content adaptive down-sampling technique using associated camera images can significantly enhance the LiDAR data compression rate with minimal impact on the accuracy of machine perception tasks. Extensive evaluations on the KITTI and Oxford datasets, using both traditional perceptual quality metrics and machine vision tasks, show that H-PCC surpasses state-of-the-art LiDAR data compression methods.

关键词： Point cloud compression Laser radar Image coding Three-dimensional displays Data compression Sensors Measurement by laser beam Octrees Entropy Image reconstruction Computer vision for automation deep learning for visual perception sensor fusion

来源：评论

学校读者我要写书评

暂无评论

Multimodal Target Localization With Landmark-Aware Positioning for Urban Mobility

引用

IEEE ROBOTICS AND AUTOMATION LETTERS 2025年第1期10卷 716-723页

作者： Hosomi, Naoki Iioka, Yui Hatanaka, Shumpei Misu, Teruhisa Yamada, Kentaro Tsukamoto, Nanami Kobayashi, Shunsuke Sugiura, Komei Keio Univ Minato Kanagawa 2238522 Japan Honda Res & Dev Co Ltd Tokyo 1076238 Japan Honda Res Inst USA Inc San Jose CA 95134 USA SOLIZE Corp Tokyo 1020075 Japan

Advancements in vehicle automation technology are expected to significantly impact how humans interact with vehicles. In this study, we propose a method to create user-friendly control interfaces for autonomous vehicles in urban environments. The proposed model predicts the vehicle's destination on the images captured by the vehicle's cameras based on high-level navigation instructions. Our data analysis found that users often specify the destination based on the relative positions of landmarks in a scene. The task is challenging because users can specify arbitrary destinations on roads, which do not have distinct visual characteristics for prediction. Thus, the model should consider relationships between landmarks and the ideal stopping position. Existing approaches only model the relationships between instructions and destinations and do not explicitly model the relative positional relationships between landmarks and destinations. To address this limitation, the proposed Target Regressor in Positioning (TRiP) model includes a novel loss function, Landmark-aware Absolute-Relative Target Position Loss, and two novel modules, Target Position Localizer and Multi-Resolution Referring Expression Comprehension Feature Extractor. To validate TRiP, we built a new dataset by extending an existing dataset of referring expression comprehension. The model was evaluated on the dataset using a standard metric, and the results showed that TRiP significantly outperformed the baseline method.

关键词： Feature extraction Natural languages Cameras Location awareness Image segmentation visualization Predictive models Urban areas Standards Optical fiber sensors deep learning methods deep learning for visual perception multi-modal perception for HRI

来源：评论

学校读者我要写书评

暂无评论

BEVCon: Advancing Bird's Eye View perception With Contrastive learning

引用

IEEE ROBOTICS AND AUTOMATION LETTERS 2025年第4期10卷 3158-3165页

作者： Leng, Ziyang Yang, Jiawei Ren, Zhicheng Zhou, Bolei Univ Calif Los Angeles Los Angeles CA 90095 USA Univ Southern Calif Los Angeles CA 90007 USA Aurora Innovat Mountain View CA 94043 USA

We present BEVCon, a simple yet effective contrastive learning framework designed to improve Bird's Eye View (BEV) perception in autonomous driving. BEV perception offers a top-down-view representation of the surrounding environment, making it crucial for 3D object detection, segmentation, and trajectory prediction tasks. While prior work has primarily focused on enhancing BEV encoders and task-specific heads, we address the underexplored potential of representation learning in BEV models. BEVCon introduces two contrastive learning modules: an instance feature contrast module for refining BEV features and a perspective view contrast module that enhances the image backbone. The dense contrastive learning designed on top of detection losses leads to improved feature representations across both the BEV encoder and the backbone. Extensive experiments on the nuScenes dataset demonstrate that BEVCon achieves consistent performance gains, achieving up to +2.4% mAP improvement over state-of-the-art baselines. Our results highlight the critical role of representation learning in BEV perception and offer a complementary avenue to conventional task-specific optimizations.

关键词： Contrastive learning Feature extraction Three-dimensional displays Transforms Representation learning Object detection Autonomous vehicles Training Annotations Head Object detection, segmentation, and categorization deep learning for visual perception representation learning

来源：评论

学校读者我要写书评

暂无评论

LBSNet: Lightweight Joint Boundary Detection and Semantic Segmentation for Transparent and Reflective Objects

引用

IEEE ROBOTICS AND AUTOMATION LETTERS 2025年第2期10卷 955-962页

作者： Tong, Ling Qian, Kun Jing, Xingshuo Southeast Univ Sch Automat Nanjing 210096 Peoples R China Minist Educ Key Lab Measurement & Control CSE Nanjing 210096 Peoples R China Southeast Univ Shenzhen Res Inst Shenzhen 518063 Peoples R China

Accurate visual detection of transparent and reflective objects remains a challenging issue for mobile manipulators. For the most common depth cameras and LiDAR sensors, the distinctive optical attributes inherent in both transparent and reflective objects pose a significant challenge. To address this problem, this study proposes a lightweight joint boundary detection and semantic segmentation network named LBSNet. LBSNet aims to enhance the perception of transparent and reflective objects in complex and dynamic environments, using RGB images only. It leverages the synergy between boundary detection and semantic segmentation through feature fusion and a multitask learning mechanism. The encoder consists of two paths: one captures category-aware semantic information, while the other discerns boundaries from multi-scale features. The gated channel adaptive (GCA) module enhances boundary features by learning channel parameters. The dynamic adaptive feature fusion (DAFF) module dynamically adjusts semantic and boundary information through cross-feature fusion. These methods effectively capture the distinctive characteristics of transparent and reflective objects, such as light refraction, boundary blurring and low contrast. Experimental results show that LBSNet achieves higher accuracy and faster processing speed on multiple public datasets compared with existing methods. Moreover, its lightweight design makes it suitable for resource-constrained mobile manipulators.

关键词： Semantic segmentation Semantics Feature extraction Accuracy Mobile robots Logic gates Computer architecture Transformers Object segmentation Computational modeling Transparent and reflective objects semantic segmentation deep learning for visual perception robot vision

来源：评论

学校读者我要写书评

暂无评论

W-ControlUDA: Weather-Controllable Diffusion-assisted Unsupervised Domain Adaptation for Semantic Segmentation

引用

IEEE ROBOTICS AND AUTOMATION LETTERS 2025年第5期10卷 4204-4211页

作者： Shen, Fengyi Zhou, Li Kuecuekaytekin, Kagan Eskandar, George Basem Fouad Liu, Ziyuan Wang, He Knoll, Alois Tech Univ Munich D-80333 Munich Germany Huawei Munich Res Ctr D-80992 Munich Germany Peking Univ EPIC Lab Beijing 100871 Peoples R China Galbot Beijing 100871 Peoples R China

Image generation has emerged as a potent strategy to enrich training data for unsupervised domain adaptation (UDA) of semantic segmentation in adverse weathers due to the scarcity of labelled target domain data. Previous UDA works commonly utilize generative adversarial networks (GANs) to translate images from the source to the target domain to enhance UDA training. However, these GANs, trained from scratch in an unpaired manner, produce sub-optimal image quality and lack multi-weather controllability. Consequently, controllable data generation for diverse weather scenarios remains underexplored. The recent strides in text-to-image diffusion models (DM) enables high fidelity diverse image generation conditioned on semantic labels. However, such DMs must be trained in a paired manner, i.e., image and label pairs, which poses huge challenge to the UDA setting where target domain labels are missing. This work addresses two key questions: What is an optimal approach to train DMs for UDA, and how can the generated data best enhance UDA performance? We introduce W-ControlUDA, a diffusion-assisted framework for UDA segmentation in adverse weather. W-ControlUDA involves two steps: DM training for data augmentation and UDA training using the generated data. Unlike previous unpaired training, our method conditions the DM on target predictions from a pre-trained segmentor, addressing the lack of target labels. We propose UDAControlNet for high-fidelity cross-domain and intra-domain data generation under adverse weathers. In UDA training, a label filtering mechanism is introduced to ensure more reliable results. W-ControlUDA helps UDA achieve a new milestone (72.8 mIoU) on the popular Cityscapes-to-ACDC benchmark and notably improves the model's generalization on 5 other benchmarks.

关键词： Training Meteorology Semantics Semantic segmentation Data collection Image synthesis Diffusion models Electronic mail Adaptation models visualization Computer vision for transportation deep learning for visual perception visual learning

来源：评论

学校读者我要写书评

暂无评论

Sparse Prototype Network for Explainable Pedestrian Behavior Prediction

引用

IEEE ROBOTICS AND AUTOMATION LETTERS 2025年第5期10卷 4196-4203页

作者： Feng, Yan Carballo, Alexander Takeda, Kazuya Nagoya Univ Grad Sch Informat Nagoya Aichi 4648601 Japan Nagoya Univ Inst Innovat Future Soc Nagoya Aichi 4648601 Japan Gifu Univ Fac Engn Grad Sch Engn Gifu 5011193 Japan Nagoya Univ Tier IV Inc Open Innovat Ctr Nagoya 4506610 Japan

Predicting pedestrian behavior is challenging yet crucial for applications such as autonomous driving and smart cities. Recent deep learning models have achieved remarkable performance in making accurate predictions, but they fail to provide explanations of their inner workings. One reason for this problem is the multi-modal inputs. To bridge this gap, we present Sparse Prototype Network (SPN), an explainable method designed to simultaneously predict a pedestrian's future action, trajectory, and pose. SPN leverages an intermediate prototype bottleneck layer to provide sample-based explanations for its predictions. The prototypes are modality-independent, meaning that they can correspond to any modality from the input. Therefore, SPN can extend to arbitrary combinations of modalities. Regularized by mono-semanticity and clustering constraints, the prototypes learn consistent and human-understandable features and achieve state-of-the-art performance on action, trajectory and pose prediction on TITAN and PIE. Finally, we propose a metric named Top-K Mono-semanticity Scale to quantitatively evaluate the explainability. Qualitative results show a positive correlation between sparsity and explainability.

关键词： Prototypes Pedestrians Trajectory Predictive models Measurement Training Vectors Semantics Encoding Explainable AI Automation technologies for smart cities deep learning for visual perception intention recognition

来源：评论

学校读者我要写书评

暂无评论

SCDA-Net: Structure Completion and Density Awareness Network for LiDAR-Based 3D Object Detection

引用

IEEE ROBOTICS AND AUTOMATION LETTERS 2025年第5期10卷 4268-4275页

作者： Wu, Shuwen Yang, Jinfu Ma, Jiaqi Zhang, Shaochen Hao, Tianhao Li, Mingai Beijing Univ Technol Sch Informat Sci & Technol Beijing 100124 Peoples R China

As a fundamental task in various application scenarios, including autonomous driving and mobile robotic systems, 3D object detection has received extensive attention from researchers in both academia and industry. However, due to the working principle of LiDAR and external factors such as occlusion, the collected point cloud of the object is usually sparse and incomplete, which affects the performance of 3D object detector. In this letter, a Structure Completion and Density Awareness Network (SCDA-Net) is proposed for 3D object detection from point clouds. Specifically, a structure completion module is designed to predict dense shapes of complete point clouds by leveraging sequence transduction ability of the transformer architecture. Furthermore, we propose a density-aware voxel RoI pooling strategy to introduce density features that reflect the state information of the original objects in refinement stage. By restoring the complete structure of the objects and considering the true distribution of the points in raw point cloud, the proposed method achieves more accurate feature extraction and scene perception. Extensive experimental results on the KITTI and Waymo datasets demonstrate the effectiveness of the proposed SCDA-Net.

关键词： Three-dimensional displays Point cloud compression Proposals Feature extraction Object detection Shape Laser radar Detectors Transformers Convolution segmentation and categorization deep learning for visual perception deep learning methods

来源：评论

学校读者我要写书评

暂无评论

DecTrain: Deciding When to Train a Monocular Depth DNN Online

引用

IEEE ROBOTICS AND AUTOMATION LETTERS 2025年第3期10卷 2822-2829页

作者： Fu, Zih-Sing Sudhakar, Soumya Karaman, Sertac Sze, Vivienne MIT Cambridge MA 02139 USA

deep neural networks (DNNs) can deteriorate in accuracy when deployment data differs from training data. While performing online training at all timesteps can improve accuracy, it is computationally expensive. We propose DecTrain, a new algorithm that decides when to train a monocular depth DNN online using self-supervision with low overhead. To make the decision at each timestep, DecTrain compares the cost of training with the predicted accuracy gain. We evaluate DecTrain on out-of-distribution data, and find DecTrain maintains accuracy compared to online training at all timesteps, while training only 44% of the time on average. We also compare the recovery of a low inference cost DNN using DecTrain and a more generalizable high inference cost DNN on various sequences. DecTrain recovers the majority (97%) of the accuracy gain of online training at all timesteps while reducing computation compared to the high inference cost DNN which recovers only 66%. With an even smaller DNN, we achieve 89% recovery while reducing computation by 56%. DecTrain enables low-cost online training for a smaller DNN to have competitive accuracy with a larger, more generalizable DNN at a lower overall computational cost.

关键词： Training Accuracy Costs Artificial neural networks Uncertainty Robots Computational efficiency Robot vision systems Depth measurement Cameras deep learning for visual perception visual learning

来源：评论

学校读者我要写书评

暂无评论

DINOv2-Based UAV visual Self-Localization in Low-Altitude Urban Environments

引用

IEEE ROBOTICS AND AUTOMATION LETTERS 2025年第2期10卷 2080-2087页

作者： Yang, Jiaqiang Qin, Danyang Tang, Huapeng Tao, Sili Bie, Haoze Ma, Lin Heilongjiang Univ Coll Elect Engn Harbin 150080 Peoples R China Southeast Univ Natl Mobile Commun Res Lab Nanjing 211102 Peoples R China Harbin Inst Technol Harbin 150080 Peoples R China

visual self-localization technology is essential for unmanned aerial vehicles (UAVs) to achieve autonomous navigation and mission execution in environments where global navigation satellite system (GNSS) signals are unavailable. This technology estimates the UAV's geographic location by performing cross-view matching between UAV and satellite images. However, significant viewpoint differences between UAV and satellite images result in poor accuracy for existing cross-view matching methods. To address this, we integrate the DINOv2 model with UAV visual localization tasks and propose a DINOv2-based UAV visual self-localization method. Considering the inherent differences between pre-trained models and cross-view matching tasks, we propose a global-local feature adaptive enhancement method (GLFA). This method leverages Transformer and multi-scale convolutions to capture long-range dependencies and local spatial information in visual images, improving the model's ability to recognize key discriminative landmarks. In addition, we propose a cross-enhancement method based on a spatial pyramid (CESP), which constructs a multi-scale spatial pyramid to cross-enhance features, effectively improving the ability of the features to perceive multi-scale spatial information. Experimental results demonstrate that the proposed method achieves impressive scores of 86.27% in R@1 and 88.87% in SDM@1 on the DenseUAV public benchmark dataset, providing a novel solution for UAV visual self-localization.

关键词： Autonomous aerial vehicles visualization Transformers Location awareness Feature extraction Satellites Adaptation models Accuracy Head Satellite images Aerial systems: Applications deep learning for visual perception localization recognition

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：