In intelligent transportation systems (ITSs), integrating pedestrians and vehicles into traffic management models is essential for developing realistic and safe solutions. However, current systems often fail to simulate complex, real-world scenarios due to the absence of a comprehensive digital twin framework across diverse traffic environments and effective modeling of pedestrian-vehicle interactions. In this article, we propose a surveillance video-assisted federated digital twin (SV-FDT) framework to enhance ITSs by incorporating pedestrians and vehicles into the control loop. SV-FDT improves computational efficiency and communication performance by transmitting only semantic data and agent parameters, rather than raw video streams. The proposed framework adopts three-layer architecture and constructs detailed pedestrian-vehicle interaction models using multi-source traffic surveillance videos. The three-layer architecture includes: (i) an end layer that collects surveillance videos from multiple sources; (ii) an edge layer that performs self-supervised semantic segmentation to extract interactions, converts them into executable traffic codes, and generates local digital twin systems (LDTSs) for regional traffic modeling; and (iii) a cloud layer that integrates LDTSs into a real-time global digital twin model. Key design considerations, challenges, and practical implementation guidelines are discussed for SV-FDT, and a testbed evaluation is used to show that SV-FDT improves traffic flow, reduces mirroring delay, and enhances recognition accuracy and system efficiency compared to traditional terminal-server frameworks. Finally, we outline open challenges and potential directions for future research in digital twin-enabled ITS.
1 June 2020 in Mechanical Systems and Signal Processing
In the last few years, enormous strides have been made for object detection and data association, which are vital subtasks for one-stage online multi-object tracking (MOT). However, the two separated submodules involved in the whole MOT pipeline are processed or optimized separately, resulting in a complex method design and requiring manual settings. In addition, few works integrate the two subtasks into a single end-to-end network to optimize the overall task. In this study, we propose an end-to-end MOT network called joint detection and association network (JDAN) that is trained and inferred in a single network. All layers in JDAN are differentiable, and can be optimized jointly to detect targets and output an association matrix for robust multi-object tracking. What’s more, we generate suitable pseudo-labels to address the data inconsistency between object detection and association. The detection and association submodules could be optimized by the composite loss function that is derived from the detection results and the generated pseudo association labels, respectively. The proposed approach is evaluated on two MOT challenge datasets, and achieves promising performance compared with classic and latest methods.
In intelligent transportation systems (ITSs), incorporating pedestrians and vehicles in-the-loop is crucial for developing realistic and safe traffic management solutions. However, there is falls short of simulating complex real-world ITS scenarios, primarily due to the lack of a digital twin implementation framework for characterizing interactions between pedestrians and vehicles at different locations in different traffic environments. In this article, we propose a surveillance video assisted federated digital twin (SV-FDT) framework to empower ITSs with pedestrians and vehicles in-the-loop. Specifically, SVFDT builds comprehensive pedestrian-vehicle interaction models by leveraging multi-source traffic surveillance videos. Its architecture consists of three layers: (i) the end layer, which collects traffic surveillance videos from multiple sources; (ii) the edge layer, responsible for semantic segmentation-based visual understanding, twin agent-based interaction modeling, and local digital twin system (LDTS) creation in local regions; and (iii) the cloud layer, which integrates LDTSs across different regions to construct a global DT model in realtime. We analyze key design requirements and challenges and present core guidelines for SVFDT's system implementation. A testbed evaluation demonstrates its effectiveness in optimizing traffic management. Comparisons with traditional terminal-server frameworks highlight SV-FDT's advantages in mirroring delays, recognition accuracy, and subjective evaluation. Finally, we identify some open challenges and discuss future research directions.
14 January 2026 in IEEE Transactions on Intelligent Transportation Systems
Digital twin (DT) technology has emerged as a key enabler for bridging the gap between physical and virtual worlds, providing real-time synchronization, monitoring, and optimization capabilities. However, in the field of intelligent transportation and autonomous driving, most existing DT systems remain limited to single-scene simulations or simplified traffic models, lacking the ability to maintain spatiotemporal consistency and continuous vehicle identities across multiple intersections. To address these limitations, this paper proposes an Urban Traffic Flow Digital Twin (UTDT) framework that builds a high-fidelity, multi-scene representation of traffic flow. UTDT integrates four core modules: single-intersection multi-object tracking, multi-intersection trajectory association, trajectory inference and restoration, and twinning synchronization. Multi-object tracking is enhanced by fusing LiDAR and camera data and leveraging re-identification for cross-scene identity continuity, enabling accurate trajectory reconstruction and precise digital twin–based vehicle control in complex traffic scenarios. The experimental results show that UTDT can provide high-fidelity simulation scenarios for the testing and optimization of autonomous driving systems, which promotes the development of intelligent transportation systems and provides important support for the construction of future smart cities. The source code and datasets are available at https://github.com/OpenHUTB/traffic_twin
Chemical short-range order (SRO), a phenomenon at the atomic scale resulting from inhomogeneities in the local chemical environment, is usually studied using machine learning force field-based molecular dynamics simulations due to the limitations of experimental methods. To promote the reliable application of machine potentials in high-entropy alloy simulations, First, this work uses NEP models trained on two different datasets to predict the Short-Range Order (SRO) coefficients of NbMoTaW. The results show that within the same machine learning framework, there are significant differences in the prediction of SRO coefficients for the Nb–Nb atomic pair. Subsequently, this work predicts the SRO coefficients of NbMoTaW using the NEP model and SNAP model, both of which are trained on the same dataset. The results revealed significant discrepancies in SRO predictions for like-element pairs (e.g. Nb–Nb and W–W) between the two potentials, despite the identical training data. The findings of this study indicate that discrepancies in the prediction results of SRO coefficients can arise from either the same machine learning framework trained on different datasets or different learning frameworks trained on the same dataset. This reflects possible incompleteness in the current training set’s coverage of local chemical environments at the atomic scale. Future research should establish unified evaluation standards to assess the capability of training sets to accurately describe complex atomic-scale behaviors such as SRO.