检索结果-内蒙古大学图书馆

arXiv 2025年

作者： Ma, Yanan Hu, Senkang Fang, Zhengru Ji, Yun Deng, Yiqin Fang, Yuguang Hong Kong JC Lab of Smart City Department of Computer Science City University of Hong Kong Hong Kong Hong Kong Key Laboratory of Advanced Sensor and Integrated System Tsinghua Shenzhen International Graduate School Tsinghua University Shenzhen518055 China

To accommodate constantly changing road conditions, real-time model training is essential for autonomous driving (AD). Federated learning (FL) serves as a promising paradigm to enable autonomous vehicles to train models collaboratively with their onboard computing resources. However, existing vehicle selection schemes for FL all assume predetermined and location-independent vehicles’ datasets, neglecting the fact that vehicles collect training data along their routes, thereby resulting in suboptimal vehicle selection. To improve the perception quality in AD for a region, we propose Sense4FL, a vehicular crowdsensing-enhanced FL framework featuring trajectory-dependent vehicular training data collection. To this end, we first derive the convergence bound of FL by considering the impact of both vehicles’ uncertain trajectories and uploading probabilities, from which we discover that minimizing the training loss is equivalent to minimizing a weighted sum of local and global earth mover’s distance (EMD) between vehicles’ collected data distribution and global data distribution. Based on this observation, we formulate the trajectory-dependent vehicle selection and data collection problem for FL in AD. Given that the problem is NP-hard, we develop an efficient algorithm to find the solution with an approximation guarantee. Extensive simulation results have demonstrated the effectiveness of our approach in improving object detection performance compared with existing benchmarks. © 2025, CC BY.

关键词： Adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

Real-time online multi-object tracking: A joint detection and tracking framework 3

Real-time online multi-object tracking: A joint detection an...

引用

3rd International Conference on Computer Science and Artificial Intelligence, CSAI 2019 and its Workshop - The 11th International Conference on Information and Multimedia Technology, ICIMT 2019

作者： Yi, Mengyun Zhang, Sheng Xu, He Department of Electronic Engineering Tsinghua University Beijing China Advanced Sensor and Integrated System Lab Tsinghua Shenzhen International Graduate School Shenzhen China

ISBN: (纸本)9781450376273

In recent years, object detection technology has been continuously developed, and the tracking-by-detection strategy has gradually become the main method of multi-object tracking. Based on detection, the accuracy of the multi-object tracking depends on the detection results to a large extent. However, in many practical applications, especially the case of complex scenes and crowded objects, the detection results are usually inaccurate. In this paper, a joint detection and tracking framework is proposed with a unified confidence scoring function to evaluate tracks confidence and complement low confidence detections with high confidence tracks. In this way, detections and tracks can be combined organically and achieved complementarity. High confidence detection results can prevent long-term tracking drift, while high confidence tracking prediction can deal with false detection and missed detection caused by occlusion during object interaction. Moreover, we trained the ReID appearance feature with higher identification capabilities on the large-scale person reidentification datasets, which has higher identification capability. Extensive experiments are conducted on MOT17 benchmarks to demonstrate the real-time and advanced performance of our tracker. © 2019 Association for Computing Machinery.

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

Real-Time Online Multi-Object Tracking: A Joint Detection and Tracking Framework 2019

Real-Time Online Multi-Object Tracking: A Joint Detection an...

引用

Proceedings of the 2019 3rd International Conference on Computer Science and Artificial Intelligence

作者： Mengyun Yi Sheng Zhang He Xu Department of Electronic Engineering Tsinghua University Beijing China Advanced Sensor and Integrated System Lab Tsinghua Shenzhen International Graduate School Shenzhen China

ISBN: (纸本)9781450376273

In recent years, object detection technology has been continuously developed, and the tracking-by-detection strategy has gradually become the main method of multi-object tracking. Based on detection, the accuracy of the multi-object tracking depends on the detection results to a large extent. However, in many practical applications, especially the case of complex scenes and crowded objects, the detection results are usually inaccurate. In this paper, a joint detection and tracking framework is proposed with a unified confidence scoring function to evaluate tracks confidence and complement low confidence detections with high confidence tracks. In this way, detections and tracks can be combined organically and achieved complementarity. High confidence detection results can prevent long-term tracking drift, while high confidence tracking prediction can deal with false detection and missed detection caused by occlusion during object interaction. Moreover, we trained the ReID appearance feature with higher identification capabilities on the large-scale person re-identification datasets, which has higher identification capability. Extensive experiments are conducted on MOT17 benchmarks to demonstrate the real-time and advanced performance of our tracker.

关键词： Joint Detection and Tracking Person Re-identification Multi-Object Tracking

来源：评论

学校读者我要写书评

暂无评论

Spatio-temporal neural network with dilated retrospective convolution for short-time change detection 7

Spatio-temporal neural network with dilated retrospective co...

引用

7th International Conference on Information, Communication and Networks, ICICN 2019

作者： Chen, Chao Zhang, Sheng Department of Electronic Engineering Tsinghua University Beijing China Advanced Sensor and Integrated System Lab Graduate School at Shenzhen Tsinghua University Shenzhen China

ISBN: (纸本)9781728104263

Change detection remains to be challenging due to variability of real-world conditions. Existing methods usually require a long-time observation, from tens to hundreds of video frames as reference, to build a robust background model for separating changing foreground from background. The performances of these methods decline severely when subject to quite fewer reference frames. In this paper, we focus on the task of short-time change detection, which aims to detect changing foreground with only a few preceding reference frames. We exploit the capability of spatial-temporal neural network in learning and representing change-cued information, and propose a novel dilated retrospective convolution which enables feature extraction between the current frame and all reference frames in a retrospective manner, and uses spatially dilated kernel for flexibly expanded inter-frame field-of-views. Trained end-to-end on hard samples from various context, our architecture performs accurate in perceiving changes and effective in combating noises in complex scenarios. Extensive evaluation on CDnet demonstrates substantial superiority of our proposed method. © 2019 IEEE.

关键词： Convolution

来源：评论

学校读者我要写书评

暂无评论

DOS-SLAM: A Real-time Dynamic Object Segmentation Visual SLAM system 19

DOS-SLAM: A Real-time Dynamic Object Segmentation Visual SLA...

引用

Proceedings of the 2019 2nd International Conference on Algorithms, Computing and Artificial Intelligence

作者： He Xu Sheng Zhang Pengyu Liu Department of Electronic Engineering Tsinghua University Beijing China Advanced Sensor and Integrated System Lab Graduate School at Shenzhen Tsinghua University Shenzhen China Institute of Microelectronics Tsinghua University Beijing China

ISBN: (纸本)9781450372619

SLAM technology has developed very rapidly in recent years, and systems for general environment has achieved excellent results. However, there are still some practical problems affecting the robustness and accuracy of SLAM, including the impact of highly dynamic environments. In this paper, we propose a real-time SLAM system, named DOS-SLAM, which is specifically designed for dynamic environments. The system, based on ORB-SLAM2, adds two modules to achieve accurate in highly dynamic environments, dynamic object segmentation with instance segmentation and geometric constraint and static semantic octo-tree map to realize a static map without moving objects. TUM RGB-D Dataset is used to present the performance of DOS-SLAM, and it has been proved that DOS-SLAM has outstanding and superior performance in dynamic environment. Meanwhile, it can realize real-time computing. In addition, static semantics octo-tree map built by DOS-SLAM system occupies less memory and only contains static objects, which is more conducive to robot navigation and large-scale scene map generation.

关键词： Visual SLAM Static semantic map Real-time Dynamic object segmentation

来源：评论

学校读者我要写书评

暂无评论

DOS-SLAM: A real-time dynamic object segmentation visual SLAM system 2

DOS-SLAM: A real-time dynamic object segmentation visual SLA...

引用

2nd International Conference on Algorithms, Computing and Artificial Intelligence, ACAI 2019

作者： Xu, He Zhang, Sheng Liu, Pengyu Department of Electronic Engineering Tsinghua University Beijing China Advanced Sensor and Integrated System Lab Graduate School at Shenzhen Tsinghua University Shenzhen China Institute of Microelectronics Tsinghua University Beijing China

ISBN: (纸本)9781450372619

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

DOS-SLAM: A Real-time Dynamic Object Segmentation Visual SLAM system

DOS-SLAM: A Real-time Dynamic Object Segmentation Visual SLA...

引用

2019第二届算法、计算和人工智能国际会议

作者： He Xu Sheng Zhang Pengyu Liu Department of Electronic Engineering Tsinghua University Advanced Sensor and Integrated System Lab Graduate School at Shenzhen Tsinghua University Institute of Microelectronics Tsinghua University

SLAM technology has developed very rapidly in recent years, and systems for general environment has achieved excellent results. However, there are still some practical problems affecting the robustness and accuracy of SLAM, including the impact of highly dynamic environments. In this paper, we propose a real-time SLAM system, named DOS-SLAM, which is specifically designed for dynamic environments. The system, based on ORBSLAM2, adds two modules to achieve accurate in highly dynamic environments, dynamic object segmentation with instance segmentation and geometric constraint and static semantic octotree map to realize a static map without moving objects. TUM RGB-D Dataset is used to present the performance of DOSSLAM, and it has been proved that DOS-SLAM has outstanding and superior performance in dynamic environment. Meanwhile, it can realize real-time computing. In addition, static semantics octotree map built by DOS-SLAM system occupies less memory and only contains static objects, which is more conducive to robot navigation and large-scale scene map generation.

关键词： Dynamic object segmentation Static semantic map Real-time Visual SLAM

来源：评论

学校读者我要写书评

暂无评论

Recent Advances in Triboelectric Nanogenerators: From Technological Progress to Commercial Applications

引用

ACS Nano 2023年第12期17卷 11087-11219页

作者： Choi, Dongwhi Lee, Younghoon Lin, Zong-Hong Cho, Sumin Kim, Miso Ao, Chi Kit Soh, Siowling Sohn, Changwan Jeong, Chang Kyu Lee, Jeongwan Lee, Minbaek Lee, Seungah Ryu, Jungho Parashar, Parag Cho, Yujang Ahn, Jaewan Kim, Il-Doo Jiang, Feng Lee, Pooi See Khandelwal, Gaurav Kim, Sang-Jae Kim, Hyun Soo Song, Hyun-Cheol Kim, Minje Nah, Junghyo Kim, Wook Menge, Habtamu Gebeyehu Park, Yong Tae Xu, Wei Hao, Jianhua Park, Hyosik Lee, Ju-Hyuck Lee, Dong-Min Kim, Sang-Woo Park, Ji Young Zhang, Haixia Zi, Yunlong Guo, Ru Cheng, Jia Yang, Ze Xie, Yannan Lee, Sangmin Chung, Jihoon Oh, Il-Kwon Kim, Ji-Seok Cheng, Tinghai Gao, Qi Cheng, Gang Gu, Guangqin Shim, Minseob Jung, Jeehoon Yun, Changwoo Zhang, Chi Liu, Guoxu Chen, Yufeng Kim, Suhan Chen, Xiangyu Hu, Jun Pu, Xiong Guo, Zi Hao Wang, Xudong Chen, Jun Xiao, Xiao Xie, Xing Jarin, Mourin Zhang, Hulin Lai, Ying-Chih He, Tianyiyi Kim, Hakjeong Park, Inkyu Ahn, Junseong Huynh, Nghia Dinh Yang, Ya Wang, Zhong Lin Baik, Jeong Min Choi, Dukhyun Kyung Hee University Gyeonggi Yongin17104 Korea Republic of Department of Electrical Engineering and Computer Science Massachusetts Institute of Technology 77 Massachusetts Avenue CambridgeMA02139 United States Department of Mechanical Engineering Soft Robotics Research Center Seoul National University Seoul08826 Korea Republic of Department of Mechanical Engineering Gachon University Seongnam13120 Korea Republic of Department of Biomedical Engineering National Taiwan University Taipei10617 Taiwan Frontier Research Center on Fundamental and Applied Sciences of Matters National Tsing Hua University Hsinchu30013 Taiwan School of Advanced Materials Science & Engineering Sungkyunkwan University Suwon16419 Korea Republic of Sungkyunkwan University 2066 Seobu-ro Gyeonggi Suwon16419 Korea Republic of Department of Chemical and Biomolecular Engineering National University of Singapore 4 Engineering Drive 4 117585 Singapore Division of Advanced Materials Engineering Jeonbuk National University 567 Baekje-daero Jeonbuk Jeonju54896 Korea Republic of Jeonbuk National University 567 Baekje-daero Jeonbuk Jeonju54896 Korea Republic of Department of Physics Inha University 100 Inha-ro Incheon22212 Korea Republic of School of Materials Science & Engineering Yeungnam University Gyeongbuk Gyeongsan38541 Korea Republic of 291 Daehak-ro Daejeon34141 Korea Republic of School of Materials Science and Engineering Nanyang Technological University 50 Nanyang Avenue 639798 Singapore Institute of Flexible Electronics Technology of Tsinghua Zhejiang Jiaxing314000 China Nanomaterials and System Lab Major of Mechatronics Engineering Faculty of Applied Energy System Jeju National University Jeju632-43 Korea Republic of School of Engineering University of Glasgow GlasgowG128QQ United Kingdom Seoul02792 Korea Republic of Department of Physics Inha University Incheon22212 Korea Republic of Suwon16419 Korea Republic of Department of

Serious climate changes and energy-related environmental problems are currently critical issues in the world. In order to reduce carbon emissions and save our environment, renewable energy harvesting technologies will serve as a key solution in the near future. Among them, triboelectric nanogenerators (TENGs), which is one of the most promising mechanical energy harvesters by means of contact electrification phenomenon, are explosively developing due to abundant wasting mechanical energy sources and a number of superior advantages in a wide availability and selection of materials, relatively simple device configurations, and low-cost processing. Significant experimental and theoretical efforts have been achieved toward understanding fundamental behaviors and a wide range of demonstrations since its report in 2012. As a result, considerable technological advancement has been exhibited and it advances the timeline of achievement in the proposed roadmap. Now, the technology has reached the stage of prototype development with verification of performance beyond the lab scale environment toward its commercialization. In this review, distinguished authors in the world worked together to summarize the state of the art in theory, materials, devices, systems, circuits, and applications in TENG fields. The great research achievements of researchers in this field around the world over the past decade are expected to play a major role in coming to fruition of unexpectedly accelerated technological advances over the next decade. © 2023 The Authors. Published by American Chemical Society

关键词： Nanogenerators

来源：评论

学校读者我要写书评

暂无评论

Clustering-Based Driving Behavior Recognition Using Inertial sensors

Clustering-Based Driving Behavior Recognition Using Inertial...

引用

International Symposium on Computational Intelligence and Design, ISCID

作者： Shuang Qin Sheng Zhang Minglin Wu Advanced Sensor and Integrated System Lab Tsinghua University Shenzhen China

ISBN: (纸本)9781538685280;9781538685273

Understanding and recognizing driving behavior is critical for traffic safety. In this paper, we propose a clustering-based systematic method to analyze and recognize driving behavior. Different from the past driving behavior recognition, we introduce wavelet analysis to get more accurately localized temporal and frequency information, and we also significantly reduce the feature dimension by the algorithm of applying metric learning to clustering. In our experiment, with only a 4-dimensional feature set, seven different driving events can be classified with a high accuracy of 97%. The proposed systematic method is universal, efficient and has a solid theoretical basis.

关键词： sensors Automobiles Acceleration Hidden Markov models Wavelet analysis Feature extraction Task analysis

来源：评论

学校读者我要写书评

暂无评论

Learning to detect instantaneous changes with retrospective convolution and static sample synthesis

arXiv

引用

arXiv 2018年

作者： Chen, Chao Zhang, Sheng Du, Cuibing Department of Electronic Engineering Tsinghua University Beijing China Advanced Sensor and Integrated System Lab Graduate School Shenzhen Tsinghua University Shenzhen China

Change detection has been a challenging visual task due to the dynamic nature of real-world scenes. Good performance of existing methods depends largely on prior background images or a long-term observation. These methods, however, suffer severe degradation when they are applied to detection of instantaneously occurred changes with only a few preceding frames provided. In this paper, we exploit spatio-temporal convolutional networks to address this challenge, and propose a novel retrospective convolution, which features efficient change information extraction between the current frame and frames from historical observation. To address the problem of foreground-specific over-fitting in learning-based methods, we further propose a data augmentation method, named static sample synthesis, to guide the network to focus on learning change-cued information rather than specific spatial features of foreground. Trained end-to-end with complex scenarios, our framework proves to be accurate in detecting instantaneous changes and robust in combating diverse noises. Extensive experiments demonstrate that our proposed method significantly outperforms existing methods. Copyright © 2018, The Authors. All rights reserved.

关键词： Convolution

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：