Unsupervised person re-identification (Re-ID) has emerged as a pivotal technology for intelligent security surveillance systems for the Internet of Things (IoT) due to its label-free deployment capability. Existing methods follow a clustering-iteration paradigm, where pseudo labels are employed to supervise Re-ID model training. However, the reliability of pseudo labels mainly depends on the quality of the learned deep embeddings, leading to the inevitable label noise. To this end, we propose a Local enhancing Window vision Transformer based on Global-Part Features (LWTGPF) to toward more challenging unsupervised Re-ID tasks. LWTGPF optimizes pseudo-label generation from both the backbone and contrastive learning. Specially, a Local enhancing window Vision Transformer (LewinVIT) is designed to serve as the backbone to learn fine-grained Re-ID representations. LewinVIT implements window-level self-attention to efficiently capture key person information. Based on this, a carefully designed camera embedding layer and a dual-branch decoupled head are integrated to achieve camera heterogeneity awareness and fine-grained feature extraction, respectively. Subsequently, we further develop a Global–Part Off/online Camera-aware Contrastive learning (GPOCC) framework. With the assistance of camera labels, GPOCC splits each single cluster into multiple proxy-level clusters. The offline association retrieves positives and negatives via clustering and splitting results. The online association dynamically associates proxies according to up-to-date proxy-level features. The association results are summarized to define global-part contrastive learning losses for end-to-end learning. GPOCC explicitly captures local structures within clusters and alleviates identity ill-posedness problem caused by the change of camera views. Extensive experiments are conducted on 5 popular person Re-ID datasets show that the proposed method achieves competitive performance and further narrows the gap between unsupervised and supervised learning. Code is available at: https://github.com/YanJieWen/LWTGPF-2025.

Simultaneous Localization and Mapping (SLAM) serves as a pivotal technology in autonomous driving. Unlike visual SLAM, which struggles in settings with inadequate visual information, LiDAR-based SLAM captures point cloud data directly, rendering a three-dimensional representation of the environment and thus offering a more robust localization capability. Given the swift advancements in autonomous driving, this work highlights the potential of combining LiDAR-based SLAM with deep learning methods, particularly in terms of improved perception and navigation within complex environments. The survey concludes by highlighting unresolved challenges and providing a practical outlook on the future of autonomous driving.

Origin–Destination (OD) prediction offers comprehensive insights into urban mobility patterns but remains challenging due to high dimensionality, data sparsity, and complex spatiotemporal dependencies. To this end, we propose a Dual-Branch Spatial–Temporal Network (DBSTNet) for forecasting urban taxi OD demand. Specially, DBSTNet models pick-up and drop-off patterns via transposed OD matrices and employs 2 independent branches to separately capture spatial–temporal features. We develop a Multi-hop Spatial-Hierarchical Temporal (MS-HT) block that leverages Chebyshev polynomial-based k-hop Graph Convolutions Networks(GCNs) to extract long-range spatial dependencies, which alleviates over-smoothing resulting from stacked GCNs. Furthermore, MS-HT also integrates Hierarchical Long Short-Term Memory (HLSTM) to learn temporal hierarchical relationships. HLSTM extends the hierarchical structure of LSTM through a nested, rather than stacked design. HLSTM unit incorporates a standard LSTM gate to manage cell state updates, enabling selective access to the internal memory. So that longer-term information which is only situationally relevant can be accessed selectively. On the NYC-Taxi, DBSTNet achieves up to a 6.05% improvement in SMAPE. On the PEMS08, the single-path version of DBSTNet outperforms state-of-the-art demand prediction models with an 8.29% increase in SMAPE, underscoring the scalability of the proposed MS-HT for spatial–temporal computation. Moreover, visualizations demonstrate that the model achieves over 85% accuracy of OD matrix, origin–destination demand prediction in most cases. Our model and dataset are open sourced at https://github.com/YanJieWen/OD-STGCN.

In tackling the intricate challenges of temporal and spatial correlation, as well as the dynamic sequential nature inherent in traffic flow data. We present a pioneering solution called the 3D Grouped Convolution Recurrent Neural Network (3D-GCRNN). The 3D convolutions adeptly capture the network’s spatial intricacies across consecutive time steps, where channel grouping ensures an efficient reduction in the computational complexity of the convolutional operations. Furthermore, the Long-Short Term Memory (LSTM) recurrent units discern temporal features characterized by long-term dependencies. Extensive empirical experiments are conducted on the Highway Gantry datasets from Fuzhou. The results shows that our 3D-GCRNN exhibits predictive performance surpassing all baselines. Additionally, the visualization results vividly illustrate the model's adeptness in capturing the evolving dynamics of traffic states accurately. This work holds great promise in enhancing traffic management efficiency and ensuring the quality of residents' travel, offering essential technological support for transportation advancements.

Conducting real-world train crash experiments is the most straightforward and effective method to research the train’s crashworthiness and enhance passive safety protection. As a non-contact measurement method, high-speed camera can efficiently capture the evolving motion patterns of trains under the high-speed states. Traditional data extraction methods rely on expert-based manual annotations, which are susceptible to factors such as illumination changes, scale variance, and shock debris. Inspired by the tremendous success of Deep Neural Networks in the computer vision community, we firstly collect 75 real-world train crash scenes and manually annotated them to form the Crash2024 dataset, enriching the community’s data resources. Moreover, we propose a novel Gradient-guided Joint representation loss with Adaptive neck Detection network (GJADet). At the macro level, we embed the adaptive module into the Path Aggregation Feature Pyramid Network, which combines multiple self-attention mechanisms to achieve scale-awareness, spatial-awareness, and task-awareness, improving the detector’s representation ability and alleviating the dense-small characteristics of the ‘point’ class without significant computational overhead. At the micro level, due to the extreme imbalance of ‘point’ class compared to other classes, we propose a gradient-guided joint representation classification loss to mitigate the long-tailed detection issue. Moreover, the classification and regression are joint representation to maintain consistency between training and inference phase. On the Crash2024, GJADet achieves the performance improvement of 3.5AP and significantly alleviate the accuracy loss problem for rare categories. Our code are open source at https://github.com/YanJieWen/GJADet-crash-pytorch.

This paper introduces FAST-LiDAR-SLAM: a robust, fast, and versatile LiDAR SLAM framework, which is specifically designed for urban scenarios with unstable GPS signals. The framework establishes a flexible and compact factor graph model that incorporates various relative and absolute measurements (including loop closure and GPS) as factors to optimize state estimation and dense mapping. Two key innovations are presented in this work. Firstly, it adopts a direct point cloud registration and map updating system in contrast to the traditional feature-based SLAM. In this way, the system can effectively exploit the fine structure of the environment and maintain robustness when confronted with non-structural road scenes. Secondly, an improved strong tracking extended Kalman filter (STEKF) is proposed to exclude anomalous GPS data in the back-end processing. Experiments show that FAST-LiDAR-SLAM achieves single-frame pose estimation in 20 $ms$ for solid-state LiDAR tests, outperforming Faster-LIO in accuracy. For rotating LiDAR, FAST-LiDAR-SLAM achieves comparable accuracy to LIO-SAM while significantly improving computational efficiency.

Short term forecasting is essential and challenging in time series data analysis for traffic flow research. A novel deep learning architecture on short-term traffic flow prediction was presented in this work. In conventional model-driven prediction method, a critical deviation in prediction accuracy was occurred in face of large fluctuations in traffic flow, while machine and deep learning-based approaches performed well in accuracy study than conventional regression-based models. Moreover, a fusion attention mechanism bidirectional long short-term memory model (ATT-BiLSTM) was proposed due to its bidirectional LSTM (BiLSTM) and attention mechanism units. The model not only dealt with forward and backward dependencies in time series data, but also integrated the attention mechanism to improve the ability on key information representation. The BiLSTM layer was exploited to capture bidirectional temporal and spatial features dependencies from historical data. The proposed model was also trained and validated using freeway toll datasets from Humen Bridge. The results showed that compared with ARIMA and SVR models, the indicators of the proposed model have been significantly improved. The ablation experiments were conducted to evaluate the role of the attention mechanism module. Compared with BiLSTM, CNN and 1DCNN-ATT-BiLSTM models, the MAE, RMSE and MAPE indexes of proposed model were reduced by 0.6–5.9%, 1.6–4.7% and 0.6–22.8%, respectively. More accurate predictions were obtained by the proposed model. The research results are of great significance to improve the level of traffic management.

This paper proposes the capacity flexibility models of transit networks to describe the ability of urban transit networks to accommodate the changes in passenger demands. In order to keep the variations of the average passenger travel time in a reasonable way, the concept of coefficient of variation (CoV) is introduced to measure and limit the deviation from a baseline time of the average passenger travel time. The CoV could transform the capacity flexibility values to non-dimensional form in order to compare the capacity flexibility characteristics of different scale transit networks on an equal basis. The genetic algorithm with deep search (GA-DS) for this problem is proposed to find the approximating optimal solution of the capacity flexibility models more effectively. The algorithm has been tested with benchmark problems reported in the existing literature. The optimal solution of the flexibility models shows that the flexibility values are further improved by reducing the number of stops for transit routes under different transit operation conditions. At last, the influence factors of the capacity flexibility of transit systems are discussed to address reliable transit services.

Rail transit plays a crucial role in improving urban sustainability and livability. In many Chinese cities, the planning of rail transit routes and stations is focused on facilitating new developments rather than revitalizing existing built-up areas. This approach reflects the local governments’ expectations of substantial growth to reshape the urban structure. However, existing research on transit-oriented development (TOD) rarely explores the spatial interactions between individual transit stations and investigates how they can be integrated to achieve synergistic effects and balanced development. This study proposes that rail transit systems impact urban structure through two “forces”: the provision of additional and reliable carrying capacity and the reduction of travel time between locations. Metro passenger flow is used as a proxy for these forces, and community detection techniques are employed to identify the actual and optimal spatial clusters in Wuhan, China. The results reveal that the planned sub-centers align reasonably well with the optimal spatial clusters in terms of spatial configuration. However, the actual spatial clusters tend to have longer internal travel times compared to the optimal clusters. Further exploration suggests the need for equalizing land use density within planned spatial clusters served by the metro system. Additionally, promoting concentrated, differentiated, and mixed functional arrangements in metro station areas with low passenger flows within the planned clusters could be beneficial. This paper presents a new framework for investigating urban spatial clusters influenced by a metro system.

结合国家综合立体交通网的建设,以城市交通站点功能组合为切入点,提出轨道交通引导城市土地开发的规划方法.以公共交通引导开发的理念为指导,面向城市群、城市与城市枢纽3类空间尺度,优化轨道交通多层级功能组织,探索站群和站点尺度一体化开发的方式方法.以厦门东站为例,依托城际铁路R1线的建设,支撑城市发展格局,明确厦门东站在翔安新城片区的功能定位,打造厦门市地标性的综合交通枢纽,统筹区域站群布局,推动轨道交通沿线的职住平衡,通过厦门东站轨道交通枢纽开发推动厦漳泉都市圈一体化发展的建议.

Traffic prediction problem is one of the essential tasks of intelligent transportation system (ITS), alleviating traffic congestion effectively and promoting the intelligent development of urban traffic. To accommodate long-range dependencies, Transformer-based methods have been used in traffic prediction tasks due to the parallelizable processing of sequences and explanation of attention matrices compared with recurrent neural units (RNNs). However, the Transformer-based model has two limitations, on the one hand, it ignores the local correlation in the traffic state in its parallel processing of the sequence, on the other hand, the absolute positional embedding is adopted to represent the positional relationship of time nodes is destroyed when it comes to calculate attention score. To address two embarrassing shortcomings, a novel framework called RPConvformer is proposed, where the improved parts are 1D causal convolutional sequence embedding and relative position encoding. In sequence embedding, we develop a sequence embedding layer composed of convolutional units, which consist of origin 1D convolutional and 1D causal convolutional. The size of the receptive field of the convolution can focus on the local region correlation of the sequence. In relative position encoding, we introduce a bias vector to automatically learn the relative position information of time nodes when linearly mapping the feature tensor. We respect the encoding and decoding framework of the Transformer, the encoder is responsible for extracting historical traffic state information, and the decoder autoregressively predicts the future traffic state. The multi-head attention mechanism is adopted by both encoder and decoder aims to focus on rich temporal feature patterns. Moreover, key mask technique is used after computing attention matrix to mask the traffic state at missing moments improving the resilience of the model. Extensive experiments on two real-world traffic flow datasets. The results show that RPConvformer achieves the best performance compared to state-of-the-art time series models. Ablation experiments show that considering the local correlation of time series has a higher gain on prediction performance. Random mask experiments show that the model is robust when the historical data is less than 10% missing. In addition, multi-head attention matrix provides further explanation for the dependence between time nodes. RPConvformer as an improved Transformer-based model can provide new ideas for molding temporal dimension in traffic prediction tasks. Our code has been open-sourced at (https://github.com/YanJieWen/RPConvformer).

In order to set reasonable prices for high speed railway（HSR） projects and achieve sustainable development of HSRs, this study proposed a pricing prediction model based on the HSR cost portrait. The model consisted of a latent factor algorithm and a convolutional neural network. In order to uncover the latent features contained in the measurable costs of HSRs, the latent factor algorithm was applied to analyze these costs. The algorithm could construct an HSR cost portrait through matrix decomposition with limited data and avoid loading sparse matrices, which effectively reduced the memory space for operating hardware. The constructed HSR cost portrait was treated as image data and input into convolutional neural networks with different architectures for training and pricing prediction. The comparison with multiple baseline models shows that multiple convolutional layers connected with pooling layers have higher accuracy in predicting HSR prices. A case study on pricing prediction for a newly built HSR project shows that the predicted results are consistent with the actual situation, which provides a reference for the research on HSR pricing prediction.

Hubs play an important role in the network of many distribution systems. However, hubs can be disrupted due to various reasons, and such disruptions can lead to a substantial transportation cost increase. In this study, we investigate the reliable multiple allocation hub location problem with fixed cost considering multiple hub disruptions. We use multiple uncertainty sets to model the uncertainty of hub disruptions and propose a model based on the two-stage robust optimization approach. To solve the two-stage robust model, we develop an efficient exact solution method based on the column-and-constraint generation algorithm. Numerical examples from the CAB dataset confirm the effectiveness and efficiency of the proposed model and the solution algorithm.

Urban public security incidents are prone to occur. Better understanding of pedestrian abnormal behavior and trajectory in crowded places is conducive to crowd management and safety monitoring. A novel pedestrian abnormal behavior detection model (PABDM) is proposed to identify crowd behavior under abnormal scenarios. This model originated from a multiscale fusion you only look once (YOLO) version 3 (V3) algorithm and was trained using the PASCAL visual object classes (VOC) in combination with an abnormal pedestrian data set (APD), denoted as VOC+APD. Compared with YOLOV3-VOC, single-stage detectors (SSD)-VOC, and SSD-VOC+APD, the proposed model has notable advantages in prediction accuracy and detection efficiency. The results show that the network loss function of the model tends to be stable after 500 epochs, and its detection accuracy is 6% higher than the average accuracy of the compared models. This proposed model also effectively solves the problem of missing detection caused by edge target, fuzzy target, and small target in abnormal state human detection. The research results are of great significance for real-time crowd monitoring in complex scenes.

交通是塑造国土空间格局、强化空间组织的重要支撑,交通规划与国土空间规划的协同发展对于优化空间格局、提升空间治理具有重要意义.政府部门是规划实施的推动者,其对交通规划与国土空间规划事权的履行情况直接影响规划的实施效果.本文梳理了现阶段交通规划与国土空间规划的事权体系,基于新发展格局中交通规划的发展重点,即满足国土空间的管控需求与实施交通强国的战略重点,分析了交通规划与国土空间规划在规划管理与实施的过程中存在的事权冲突,主要包括:保护与发展、短期与长期的目标冲突,规划管理、传导方面的管理冲突,法律地位缺失、权限模糊的法治冲突.针对存在的冲突并考虑未来的发展重点,提出交通规划与国土空间规划的协同发展思路,其关键在于管理部门间具有目标认可、法律保障、沟通合作,具体来说是坚持统筹协调,强化目标融合;完善法律体系,明确权责边界;推进多方联动,健全协作机制.

PurposeIdentifying the complementary effects of ride-sharing on public transit is critical to understanding the potential value of growing partnerships between public transit agencies and ride-sharing platforms. The purpose of this paper is to investigate whether and how ride-sharing services complement public transit.Design/methodology/approachTaking advantage of a natural experiment whereby subway Line 2 opened after the entry of ride-sharing services in Xiamen, this study uses a difference-in-differences approach to identify the complementary effects of ride-sharing on public transit based on a proprietary fine-grained trip-level data set from a large ride-sharing platform.FindingsThis study obtained the encouraging finding that ride-sharing has a significant complementary effect on the subway, as the number of ride-sharing pickups and drop-offs at subway stations increased by 130% and 117.9%, respectively, after the subway opening. Moreover, mechanism analysis shows that the complementary effect of ride-sharing services is stronger when connection distance is short (i.e. under 6 km) and when the transportation availability is limited (i.e. at night or in the areas with low transit supply and low population density).Practical implicationsThe findings provide guidelines for promoting cooperation between public transit agencies and ride-sharing platforms to build an efficient and sustainable urban transport system.Originality/valueThis study is the first to examine the complementary effect of ride-sharing services on public transit via unique fine-grained ride-sharing trips data, and further reveal the underlying mechanism behind this effect.

针对是否能够通过改变建成环境改善高收入群体对小汽车出行模式的依赖,利用厦门市2015年居民出行调查数据,基于多项Logit模型,探究了建成环境对有车/无车高收入群体出行模式影响的差异性,探究了个人社会经济属性和建成环境对出行模式的影响强度,并分析了对高收入群体出行模式影响显著的关键因子及其影响机制.结果表明:在控制其他变量后,交通小区人口密度与土地利用混合度的提高抑制了高收入群体对小汽车的使用,但是抑制作用较弱;就业密度和公交站点密度与有车高收入群体的出行模式无显著关联;交通小区路网密度、停车位密度、绿化率的提高促进了拥有小汽车的高收入群体使用小汽车;交通小区购物店密度和休闲娱乐场所密度的提高促进了高收入群体的步行水平;交通小区容积率与高收入群体的出行模式无显著关联;厦门岛内高收入群体在出行时更多选择步行与公共交通,岛外高收入群体则更多倾向选择小汽车;建成环境对有车/无车的高收入群体出行模式的影响呈现出显著差异性,其作用程度小于个人社会经济属性对出行模式的影响.以上结论可为通过优化不同类群居民居住区域的土地利用规划用以改善居民出行模式和城市规划提供参考.

In this study, the risk and benefit mechanism of an information integration system embedded into a system dynamics model is applied to calculate the overall risk of urban rail transit public–private partnership (PPP) projects. Moreover, the cash flow and accumulation of a PPP project are simulated. In the risk mechanism information integration system, a risk allocation decision model is established through an empirical case study, and a correlation analysis is performed on the evaluation of different risk allocation schemes to determine the best decision scheme for each risk factor, social capital, and the government. In the return mechanism information integration process, the final decision-making scheme is obtained from the perspective of social capital, the government, and the public through the analysis of the net present value. Chengdu Rail Transit Line 17 phase I PPP project is presented as an empirical case. The risk allocation and return mechanism information integration scheme is determined by applying the methods in detail.

为了解决交通信用评级问题,基于组合客观赋权法对交通运输企业进行了合理评级.分析了交通信用评级指标的构成;利用组合客观赋权法结合连续式3-sigma准则获得了交通运输企业的信用得分,并采用区间式3-sigma准则对交通运输企业信用进行评级;以厦门市各运输企业的信用相关指标数据为例进行算法性能分析.研究结果表明:企业监督检查(cs)权重最大;对于企业信誉考评指标而言,其权重随着信誉等级的下降而降低.

针对目前我国城市网约车驾驶员分类管理的难题,首先分析了国内外网约车平台和典型城市网约车驾驶员分类管理机制,基于此提出了我国城市网约车驾驶员分类管理的原则、操作方法、驾驶员考核和准入机制;最后提出了专、兼职网约车驾驶员分类管理的服务基础.基于目前我国城市网约车管理实行"遵上位法、采用一刀切、不分类、不允许兼职"的驾驶员管理模式,本文提出了平台约束法、总量控制法、"潮汐"运力模型法和全有全无法等丰富我国城市网约车驾驶员的分类管理方法.研究成果可为我国城市网约车运营管理标准化提供决策支持.