检索结果-内蒙古大学图书馆

IEEE International Conference on Real-time Computing and Robotics (RCAR)

作者： Guanfeng Yu Han Zhang Zehao Wang Jingchuan Wang Department of Automation Institute of Medical Robotics Shanghai Jiao Tong University Key Laboratory of System Control and Information Processing Ministry of Education of China Shanghai Engineering Research Center of Intelligent Control and Management Shanghai China

ISBN: (数字)9798350372601

ISBN: (纸本)9798350372618

The paper introduces a Koopman bilinear model predictive control (KBMPC) as a local planner for smart wheelchair systems. This approach leverages the Koopman Operator to streamline the nonlinear terms in MPC, aiming to achieve superior real-time performance. In particular, we employ the Extended Dynamic Mode Decomposition (EDMD) method to approximate the infinite dimensional Koopman operator. To tackle the challenge of designing lifting functions, we systematically utilize the derivatives of known nonlinear structures to methodically derive lifting functions. This systematic approach facilitates the identification of a precise bilinear model. Next, we formulate the corresponding Koopman bilinear MPC problem. Finally, an iterative algorithm is employed to efficiently solve the KBMPC problem. This method linearizes the bilinear model around estimations and iteratively solves the resulting quadratic programming problem to refine estimations. This allows KBMPC to function as a real-time local planner for smart wheelchairs. The prediction of nonlinear terms shows that the identified Koopman bilinear model increases the prediction accuracy. Furthermore, experiments demonstrate the advantages of the designed KBMPC local planner, highlighting its superior real-time performance and predictive capabilities.

关键词： Pedestrians Accuracy systematics Wheelchairs Estimation Predictive models Real-time systems

来源：评论

学校读者我要写书评

暂无评论

GOGICP: A Real-time Gaussian Octree-based GICP Method for Faster Point Cloud Registration

GOGICP: A Real-time Gaussian Octree-based GICP Method for Fa...

引用

2024 IEEE International Conference on Real-Time Computing and Robotics, RCAR 2024

作者： Yu, Zhitao Yuan, Wei Zhao, Hengwang Zhuang, Hanyang Yang, Ming Shanghai Jiao Tong University Department of Automation Shanghai200240 China Ministry of Education of China Key Laboratory of System Control and Information Processing Shanghai200240 China Shanghai Jiao Tong University Global Institute of Future Technology Shanghai200240 China University of Michigan Shanghai Jiao Tong University Joint Institute Shanghai Jiao Tong University Shanghai200240 China

ISBN: (纸本)9798350372601

Point cloud registration algorithms are crucial for robot localization based on prior maps. Classical methods often struggle to achieve both high accuracy and real-time performance simultaneously, as these two goals can conflict. This limitation impedes the application of point cloud registration in high-speed scenarios, such as the LiDAR localization of fast-moving robots, where both precision and real-time capabilities are essential. This paper proposes a real-time Gaussian Octree-based Generalized Iterative Closest Point (GICP) method, which leverages the censoring properties of the Gaussian Octree to expedite the nearest neighbor search and incorporates an incremental calculation of the Gaussian distribution to hasten the tree-building process. Experimental results show that the proposed method can achieve a 61.8% improvement in speed over GICP while maintaining accuracy comparable to that of GICP. The average registration accuracy is up to 5 cm, with an angular error of less than 0.2°, and the average registration time for a single frame is 77 ms, which satisfies the demands of practical application scenarios. © 2024 IEEE.

关键词： Gaussian distribution

来源：评论

学校读者我要写书评

暂无评论

DELFlow: Dense Efficient Learning of Scene Flow for Large-Scale Point Clouds

arXiv

引用

arXiv 2023年

作者： Peng, Chensheng Wang, Guangming Lo, Xian Wan Wu, Xinrui Xu, Chenfeng Tomizuka, Masayoshi Zhan, Wei Wang, Hesheng Department of Automation Key Laboratory of System Control and Information Processing of Ministry of Education Shanghai Jiao Tong University China Mechanical Systems Control Laboratory University of California Berkeley United States

Point clouds are naturally sparse, while image pixels are dense. The inconsistency limits feature fusion from both modalities for point-wise scene flow estimation. Previous methods rarely predict scene flow from the entire point clouds of the scene with one-time inference due to the memory inefficiency and heavy overhead from distance calculation and sorting involved in commonly used farthest point sampling, KNN, and ball query algorithms for local feature aggregation. To mitigate these issues in scene flow learning, we regularize raw points to a dense format by storing 3D coordinates in 2D grids. Unlike the sampling operation commonly used in existing works, the dense 2D representation 1) preserves most points in the given scene, 2) brings in a significant boost of efficiency, and 3) eliminates the density gap between points and pixels, allowing us to perform effective feature fusion. We also present a novel warping projection technique to alleviate the information loss problem resulting from the fact that multiple points could be mapped into one grid during projection when computing cost volume. Sufficient experiments demonstrate the efficiency and effectiveness of our method, outperforming the prior-arts on the FlyingThings3D and KITTI dataset. Our source codes will be released on https://***/IRMVLab/DELFlow. © 2023, CC BY.

关键词： Efficiency

来源：评论

学校读者我要写书评

暂无评论

DSLO: Deep Sequence LiDAR Odometry Based on Inconsistent Spatio-temporal Propagation

DSLO: Deep Sequence LiDAR Odometry Based on Inconsistent Spa...

引用

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

作者： Huixin Zhang Guangming Wang Xinrui Wu Chenfeng Xu Mingyu Ding Masayoshi Tomizuka Wei Zhan Hesheng Wang Department of Automation Key Laboratory of System Control and Information Processing of Ministry of Education Key Laboratory of Marine Intelligent Equipment and System of Ministry of Education Shanghai Engineering Research Center of Intelligent Control and Management Shanghai Jiao Tong University Shanghai China Department of Engineering University of Cambridge Cambridge U.K. UC Berkeley Berkeley CA USA

ISBN: (数字)9798350377705

ISBN: (纸本)9798350377712

This paper introduces a 3D point cloud sequence learning model based on inconsistent spatio-temporal propagation for LiDAR odometry, termed DSLO. It consists of a pyramid structure with a spatial information reuse strategy, a sequential pose initialization module, a gated hierarchical pose refinement module, and a temporal feature propagation module. First, spatial features are encoded using a point feature pyramid, with features reused in successive pose estimations to reduce computational overhead. Second, a sequential pose initialization method is introduced, leveraging the high-frequency sampling characteristic of LiDAR to initialize the LiDAR pose. Then, a gated hierarchical pose refinement mechanism refines poses from coarse to fine by selectively retaining or discarding motion information from different layers based on gate estimations. Finally, temporal feature propagation is proposed to incorporate the historical motion information from point cloud sequences, and address the spatial inconsistency issue when transmitting motion information embedded in point clouds between frames. Experimental results on the KITTI odometry dataset and Argoverse dataset demonstrate that DSLO outperforms state-of-the-art methods, achieving at least a 15.67% improvement on RTE and a 12.64% improvement on RRE, while also achieving a 34.69% reduction in runtime compared to baseline methods. Our implementation will be available at https://***/IRMVLab/DSLO.

关键词： Point cloud compression Solid modeling Laser radar Accuracy Three-dimensional displays Runtime Logic gates Robot sensing systems Real-time systems Odometry

来源：评论

学校读者我要写书评

暂无评论

Trajectory Optimization for Multi-Lane Platoon Formation with Undefined Configurations

Trajectory Optimization for Multi-Lane Platoon Formation wit...

引用

International Conference on Intelligent Transportation

作者： Siwen Yang Yunwen Xu Dewei Li Chen Zhang School of Electronic Information and Electrical Engineering Shanghai Jiao Tong University Shanghai China Key Laboratory of System Control and Information Processing Ministry of Education of China Shanghai China

Many studies have demonstrated the potential benefits of connected and autonomous vehicles (CAVs) platooning to energy consumption and traffic efficiency. However, how to implement platoon formation, which assigns and regulates vehicles scattered in different lanes to platoons, while reducing vehicle delay and time consumption remains an open issue. Besides, most existing studies require a predetermined platoon configuration, which decreases maneuverability. This paper proposes a multilane platoon formation strategy with undefined configurations. The vehicle sequence in newly formed platoons along with the longitudinal trajectories of all vehicles before changing lanes is optimized in a nonlinear programming model. The multi-objective function considers the average velocity of all CAVs and the time consumption of formation. A Fibonacci search algorithm with an embedded mixed integer linear programming (MILP) model is developed to obtain solutions efficiently. The numerical experiments indicate the superiority of the proposed method in optimizing the vehicles' trajectories and improving multi-lane platoon formation efficiency.

关键词：

来源：评论

学校读者我要写书评

暂无评论

LHMap-loc: Cross-Modal Monocular Localization Using LiDAR Point Cloud Heat Map

arXiv

引用

arXiv 2024年

作者： Wu, Xinrui Xu, Jianbo Hu, Puyuan Wang, Guangming Wang, Hesheng Department of Automation Key Laboratory of System Control and Information Processing of Ministry of Education Key Laboratory of Marine Intelligent Equipment and System of Ministry of Education Shanghai Engineering Research Center of Intelligent Control and Management Shanghai Jiao Tong University Shanghai200240 China Department of Engineering University of Cambridge CambridgeCB2 1PZ United Kingdom

— Localization using a monocular camera in the pre-built LiDAR point cloud map has drawn increasing attention in the field of autonomous driving and mobile robotics. However, there are still many challenges (e.g. difficulties of map storage, poor localization robustness in large scenes) in accurately and efficiently implementing cross-modal localization. To solve these problems, a novel pipeline termed LHMap-loc is proposed, which achieves accurate and efficient monocular localization in LiDAR maps. Firstly, feature encoding is carried out on the original LiDAR point cloud map by generating offline heat point clouds, by which the size of the original LiDAR map is compressed. Then, an end-to-end online pose regression network is designed based on optical flow estimation and spatial attention to achieve real-time monocular visual localization in a pre-built map. In addition, a series of experiments have been conducted to prove the effectiveness of the proposed method. Our code is available at: https://***/IRMVLab/LHMaploc. Copyright © 2024, The Authors. All rights reserved.

关键词： Optical radar

来源：评论

学校读者我要写书评

暂无评论

Learning-Based Perimeter Invasion Design with Partial information of Defense Model

引用

IFAC-PapersOnLine 2023年第2期56卷 1319-1324页

作者： Yan Zhong Yushan Li Xiaopeng Wang Chongrong Fang Jianping He The Dept. of Automation Shanghai Jiao Tong University and Key Laboratory of System Control and Information Processing Ministry of Education of China Shanghai China NSFOCUS Technologies Group Co. Ltd Beijing China

This paper focuses on the learning-based perimeter-defense problem. Specifically, we consider a scenario where an attacker invades an area protected by a defender with only partial information about the target area and defense strategy. The attack design is challenging since the flexible and unknown defense strategy results in the highly uncertain feasible invasion space. To address the problem, we propose a learning-based method by using patrol and defense trajectories. First, we apply an ellipse fitting method to regress the perimeter of the protected area with piecewise elliptic segments. Then, we characterize the defense behaviors into different patterns and learn the conditions to activate different strategies by tentative invasions. Finally, we design a model predictive controller to solve the optimal invasion trajectory planning. Simulations are provided to illustrate the feasibility and effectiveness of the proposed method.

关键词： Perimeter-defense learning model-predictive control control design

来源：评论

学校读者我要写书评

暂无评论

Smoke Detection In Complex Scenery using Improved YOLOV5

Smoke Detection In Complex Scenery using Improved YOLOV5

引用

6th International Conference on Intelligent Computing and Signal processing (ICSP)

作者： ZiJun Nie YongQi Hu ZiZheng Yu RuJia Li ZhiJian Yin HuaLin Zhan Zhen Yang School of Communications and Electronics Jiangxi Science and Technology Normal University Nanchang China Key Laboratory of System Control and Information Processing Ministry of Education Shanghai China

Considering smoke detection in a complex scene, high smoke object error rate and low detection efficiency, an improved YOLOV5 for smoke detection is proposed in this paper. To improve the effectiveness of smoke detection algorithm for smoke targets in complex scenes. First, a new lightweight convolution technique GSConv is introduced to replace the Conv layer in the neck layer for feature extraction, and the original C3 module is replaced by Cross Stage Partial Network (GSCSP) module VoVGSCSP. This improvement reduces the computation complexity of the algorithm while maintaining the detection accuracy. Second, CIoU is replaced by SIoU as the regression function of the prediction box to improve the prediction accuracy of the prediction box and reduce the rate of missed detection of the target. Finally, to improve the performance of the deep network, Dynamic ReLU is introduced as an activation function (dynamically adjusting the ReLU parameters depending on the input data). Experiments on the SM-dataset (smoke detection dataset) show that the improved model improves the accuracy and precision of target smoke detection compared to the YOLOV5 model. Its precision is increased by 1.8%, mAP@0.5 is increased by 1.0%, the complexity of the model is reduced by 23.8%. The improved algorithm proposed in this paper can effectively extract smoke features and more suitable for smoke detection tasks in complex scenes.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Dynamic Template Update Mechanism for Visual Object Tracking

Dynamic Template Update Mechanism for Visual Object Tracking

引用

6th International Conference on Intelligent Computing and Signal processing (ICSP)

作者： YongQi Hu ZhongWang Cai ZiJun Nie ZiZheng Yu Ying Cao Zhijian Yin Zhen Yang School of Communications and Electronics Jiangxi Science and Technology Normal University Nanchang China Key Laboratory of System Control and Information Processing Ministry of Education Shanghai China

Given the requirements for robust target classification and accurate target state estimation in visual tracking, SiamFC++ proposes a set of practical guidelines for designing high-performance general-purpose trackers by considering the special nature of visual tracking problems. Inspired by dynamic modules, We propose an empirical method for integrating a dynamic module into the image input, which is concatenated with the template module after feature maps are extracted by the backbone network. Since the position and shape of the object can change significantly within a video sequence, the added dynamic module can better focus on the target region of the feature map to obtain better similarity maps. Extensive experiments and comparisons demonstrate that our simple and effective method achieves reliable results on the benchmarks of LaSOT, TrackingNet, and GOT-10K and provides a significant speed advantage in real-time.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Weed detection based on improved Yolov5

Weed detection based on improved Yolov5

引用

6th International Conference on Intelligent Computing and Signal processing (ICSP)

作者： ZiZheng Yu ZiJun Nie Yong Qi Hu HuangBin Yan LiJun Lou Zhijian Yin Zhen Yang School of Communications and Electronics Jiangxi Science and Technology Normal University Nanchang China Key Laboratory of System Control and Information Processing Ministry of Education Shanghai China

The scale of grain production affects human life and development. With the continuous expansion of cultivated land, the reproductive ability of weeds to mutiply gradually increases, which affects the growth of crops. If weeds are not treated properly managed, they can also allow spread indiscriminately in the field and reduce crop yields. The growth ability of weeds is stronger than that of crops, and there are many types of weeds with a wide of species diversity. To solve the problem that the existing weed detection methods cannot detect and classify weeds accurately and quickly, a deep learning method based on improved Yolov5 was designed for weed detection. By replacing the 3 × 3 convolution with multi-head self-attention (MHSA) in the Yolov5's backbone, the accuracy of weed detection is improved. The experimental results show that the improved Yolov5 weed detection algorithm achieves 51.4% accuracy. Compared with the original Yolov5 model, the calculation amount is 3 points less than the original, and total time required is slightly shorter, which improves the accuracy of weed classification and positioning.

关键词：

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：