Fully-sampled vehicle trajectories provide a complete picture of the traffic flow, which is essential for many Intelligent Transportation System (ITS) applications. Due to the high cost of vehicle trajectories collection, trajectory reconstruction issue has attracted a lot of interest. However, most methods assume a First-In-First-Out (FIFO) principle, limiting their accuracy and applicability to multilane roads. This study proposes an integrated reconstruction framework that simultaneously considers car-following (CF) and lane-changing (LC) behaviors, by fully exploiting observations from roadside sensors. Specifically, a probabilistic LC point identification model is first developed, which formulates LC occurrences as the joint outcome of motivation and feasibility. The identified LC points divide each vehicle’s motion into multiple segments, within which CF behavior dominates. To approximate these CF dynamics, a bidirectional reference points generation algorithm is then proposed, providing reliable intermediate kinematic constraints and effectively mitigating cumulative errors. Finally, within a unified optimization framework, vehicle trajectories are reconstructed by minimizing discrepancies between estimated and reference points while maintaining consistency with macroscopic traffic flow patterns. The proposed method is evaluated using the NGSIM and HighD datasets. Results show strong consistency with ground-truth trajectories under both free-flow and congested conditions, outperforming baselines in terms of location and speed accuracy. LC behaviors are also effectively reproduced, thereby improving the overall accuracy and reliability of trajectory reconstruction. Moreover, the computational efficiency indicates the promising prospects for large-scale data processing and real-time applications.

Accurate traffic prediction is crucial for the management of intelligent transportation systems. Recently, researchers have developed spatiotemporal graph neural networks (ST-GNNs), achieving significant progress in performance. However, most ST-GNNs construct the graph adjacency matrix using predefined rules or trainable parameters, without fully leveraging the congestion relationships within traffic flows to guide graph structure learning. Addressing this issue, we propose a dynamic spatiotemporal graph convolutional network (DSTGCN), which adequately considers the impact of traffic congestion on traffic prediction. Firstly, based on the fundamental traffic flow graph model and congestion propagation path algorithm, a congestion propagation matrix among road nodes is constructed. Then, the real-time congestion propagation matrix is fused with the static road adjacency matrix to establish a time-varying dynamic graph neural network structure, enabling dynamic capture of traffic state changes. Finally, complex spatiotemporal correlations are captured through a specially designed spatiotemporal convolutional architecture for multi-horizon traffic prediction. Extensive experiments conducted on real-world datasets demonstrate the effectiveness and interpretability of our approach.

To address the collaborative issue in large-scale urban rail transit (URT) network operations, this paper proposes an adaptive real-time control framework based on the Soft Actor-Critic (SAC) deep reinforcement learning (DRL) method, featuring flexible train scheduling capabilities. First, by analyzing dynamic passenger travel behavior (e.g., entering/exiting stations, transferring) and train operation events (e.g., dispatching, interstation running, station dwelling), the control problem is modeled as a Markov Decision Process (MDP) and an efficient URT simulation environment is constructed. Then, considering constraints such as train capacity and dispatch intervals, a train scheduling model is developed to minimize both passenger costs and operational costs. Subsequently, the real-time state of the URT system is represented by the overall number of passengers present at every platform, and train dispatch intervals on all lines are used as decision variables. A solving algorithm based on the SAC framework is developed. Finally, experimental results on a large-scale URT network comprising 10 lines demonstrate the effectiveness of the proposed framework, showing superior performance compared to other reinforcement learning algorithms and traditional heuristic optimization algorithms. The proposed approach achieves a 1.63% reduction in average passenger waiting time, equivalent to 2.09 seconds, while utilizing 49 fewer trains, representing a 2.97% decrease, compared to the second-best TD3 algorithm.

Given that the standardization of automated driving scenario testing is currently underway in various regions, this paper focused on the pedestrian crossing case outlined in a recent proposed international standard for terminology definitions. This study extracted human skeletal information from the Joint Attention in Autonomous Driving (JAAD) dataset, serving as a straightforward means to extract posture features. After defining a novel array of static and dynamic skeletal features, we conducted a comparative analysis of four models across two scenarios: pedestrians walking individually or in groups. To gain a more comprehensive understanding of the underlying mechanisms influencing pedestrian crossing decisions, we employed SHapley Additive exPlanation (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) to provide precise insights into both global and local predictions. The results show that heightened willingness to urgently cross the street is correlated with more noticeable knee flexion and leg alternation, a narrower shoulder in sight, and larger strides. Conversely, frequent body rotation may suggest a temporary reluctance to cross. Additionally, there is an indication that pedestrian crossing intention can be influenced by group size, vehicle movement and contextual features. Finally, practical suggestions based on our results are provided for automated driving scenario testing.

Most of the existing map matching methods are developed considering global positioning system (GPS) data. However, the limitations of GPS data, such as a small sample size and presence of position errors, impede the application of such data in intelligent transportation system (ITS) applications. In contrast, automatic vehicle identification (AVI) data exhibit unique advantages and can overcome the issues related to the GPS data. The existing map matching methods are unsuitable for use with AVI data due to the sparsity issue. Therefore, in this study, a map matching method named AVI-MM is developed that is customized for sparse AVI data. The AVI-MM first decomposes the AVI trajectories with the connective observation pairs. Subsequently, the set of candidate sub-paths for each sensor pair is generated, and each candidate sub-path is assigned a matching probability. Finally, the matching path is determined by connecting the sub-path with the highest matching probability. As many sub-paths can connect the sparse AVI observations, a candidate set generation algorithm is developed to generate a sufficient number of feasible and attractive candidates. Moreover, to reliably and accurately determine the true path, the matching probability is defined based on a spatial-temporal analysis and drivers’ route choice behaviour analysis. Field-test data are used to estimate and evaluate the proposed method. Compared with three benchmark methods, the proposed AVI-MM exhibits a significantly enhanced matching accuracy without a considerable loss in the computation efficiency. Moreover, the evaluation results of the robustness of the proposed algorithm suggest that the algorithm can correctly identify 82.55% of the links when the spatial gap between the observation pair is less than 20 km. Specifically, the proposed method exhibits a satisfactory performance for 84.02% of the observed samples in the test dataset.

In the MaaS (Mobility as a Service) age, the alternatives of route choice for a trip will not be the single mode paths, but the combined routes utilizing more than one travel mode in the multi-modal transportation systems. When modeling the multi-modal combined route choices, alternative routes are correlated not only because of the overlapping of physical links, but also because of the overlapping of travel modes. We define this problem as the generalized path overlapping problem. To address the generalized overlapping problem, a multi-modal logit kernel (MLK) model is proposed to explicitly consider the correlations of unobserved utilities of combined routes. In this model, the unobserved utilities of combined routes are divided to two parts, the link specific parts and the independent route specific parts. The link specific parts are further divided to two parts: physical link specific and the mode specific. The generalized path overlapping problem is captured by the sharing of these two parts of unobserved link utilities. Based on this model, the stochastic user equilibrium on multi-model transport networks is represented as a fixed-point problem. Numerical studies are designed to illustrate the effects of incorporating generalized overlapping problem on disaggregated route choice prediction and aggregated traffic flow assignment. With different settings of physical link and mode specific error terms, the variations of disaggregated route choice predictions and aggregated traffic flows are discussed. It is found that, the generalized overlapping problem can be captured by the proposed model and significantly affect both the individual route choice predictions and aggregated traffic flows at equilibrium.