The connected vehicles have shown great potential in improving traffic efficiency, but their influence on mixed traffic has not been fully understood. This study proposed and evaluated a simple cooperative control method at unsignalized intersections. All vehicles in this study were assumed to be connected, including connected autonomous vehicles and connected manual vehicles. Both of them could receive real-time information and instructions from control center. Vehicles without conflict relationships were allowed to enter an intersection simultaneously by following the suggested order. Moreover, overtaking was allowed based on the predicted probability of success, which leaded to a high level of vehicle cooperation. A microsimulation was developed and implemented based on cellular automata to validate the proposed method. By comparing with priority control and signal control, the proposed method was demonstrated efficient under traffic flow rates that did not cause congestion, and resulted in better efficiency when the penetration of automated vehicles exceeds 75
Urban corridors in large Indian cities continue to face severe congestion as traffic demand grows faster than available roadway capacity. The Fateh Nagar Railway Flyover in Hyderabad is one such location, routinely carrying flows well above 3,200 PCU/hr, far beyond the Indo-HCM guideline for a single lane. Under these conditions, vehicles crawl at 6–12 km/h and experience delays of more than 200 s, indicating a failing level of service. To examine these conditions and assess possible remedies, this study brings together three commonly used but rarely integrated tools: the BPR speed–flow model, PTV VISSIM microsimulation, and an Artificial Neural Network-based delay estimator. The workflow involved calibrating the BPR parameters, tuning driver-behaviour settings in VISSIM to reflect the observed mixed-traffic environment, and validating the predictions using both regression models and the neural network. Model performance was evaluated with standard statistical measures. Improvement scenarios primarily involved widening the Balkampet–Fateh Nagar section and introducing a parallel Railway Over Bridge (ROB), which were then tested. The calibrated BPR model reproduced field delays with errors under four seconds, while VISSIM reflected actual speeds and flows within a 10
Intersections on high-speed corridors carrying heterogeneous traffic often become critical blackspots due to inadequate control and geometric deficiencies. This study investigates safety and operational issues at an unsignalized X-type intersection where a Major District Road (MDR) intersects the NH66 in Thiruvananthapuram District, Kerala. Field data comprising 24-hour classified traffic counts, spot-speed measurements, detailed geometric surveys, and three-year crash records were collected and analysed. Capacity analysis showed major-road approaches operating at 49–82
Despite great advances in object detection approaches based on deep learning, current approaches have often struggled to balance high accuracy, high inference speed, and the capacity to operate in unbalanced environments, limiting system deployment in edge computing and Internet-of-Things (IoT) applications. In response, the present investigation presents a lightweight ensemble framework that is based on YOLOv8n architectures and its hyperparameters are updated using a Newton-Raphson inspired strategy. Two constituent models, instantiated at input resolutions of 320 × 320 and 640 × 640 pixels, have been trained and tested on the PKLot benchmark (containing 12,416 images) with further robustness tests done under regard of such simulated meteorological phenomena as rainy, foggy, motion-blurry, and low-light scenarios. The findings show that both configurations of the models achieve stable F1-scores of around 0.664, and the higher resolution variant gets superior localization fidelity (mAP@0.5 = 0.9891) without the cost of additional runtime. End-to-end inference rates exceed 250 frames per second and therefore meet the real-time operation requirements. Robustness analyses support high reliability under rain, fog, and motion blur although performance under low-light is less than ideal. The proposed framework delivers a deployment-ready smart parking solution, ensuring high precision and flexible operation across heterogeneous edge and IoT hardware platforms.
Road disruptions due to work zones or other factors often cause lane closures and modifications, which intensify congestion and increase the risk of accidents. By utilising Connected Automated Vehicles (CAVs) capabilities, enhanced strategies can be implemented to handle these situations more safely and efficiently. Methods proposed by the literature work under restrictive assumptions, such as specific traffic demand and static and predefined control zones, that make them less suitable for real-world conditions with varying levels of traffic demand. The ADAPTMET algorithm is designed to address these gaps by dynamically adapting the length of metering zones based on real-time traffic flow within the work zone. The function of these metering zones is the longitudinal control of the vehicles in terms of the gap between them for better merging into the open lane. This adaptive control strategy optimizes traffic flow by decreasing the control length to remove unnecessary control in uncongested states. Contrastingly, during congested states, it extends the control length to manage higher traffic volumes better, ensuring safer and more orderly movement through the work zone. Simulation-based evaluation of ADAPTMET shows throughput improvements ranging from 1
Road administrators widely rely on visual monitoring of road videos to detect vehicles with abnormal behavior, such as vehicles involved in accidents and broken-down vehicles, at an early stage. However, continuously monitoring video from all road surveillance cameras with limited personnel is impractical. Therefore, machine learning-based methods for detecting abnormal vehicle behavior from road videos have been studied to improve the efficiency of road management. However, many existing methods require a sufficient amount of labeled data for training, which makes the construction of classifiers highly time-consuming and labor-intensive. To address this issue, this paper proposes a method that uses all vehicles captured during periods in which no vehicles with clearly abnormal behavior are recorded as training data, without vehicle-level checking or labeling. Assuming that most vehicles in these periods exhibit typical behavior, the method mainly learns their feature distribution and classifies vehicles that deviate from it as “vehicles with unusual behavior.” In practical road management, the proposed method is expected to improve monitoring efficiency by enabling road administrators to focus on checking the detected vehicles with unusual behavior. Finally, experiments using road videos from ordinary roads were conducted to verify the effectiveness of the proposed method.
Traffic congestion in urban areas remains a major challenge in the context of smart cities, primarily due to suboptimal awareness mechanisms and unreliable rerouting plans in traditional VANET-based Intelligent Transportation Systems. This paper proposes a congestion-aware traffic control framework, ICDDR (Intelligent Congestion Detection, Dissemination, and Rerouting), which utilizes a unified congestion detection (CD) score to jointly regulate alert propagation range and vehicle rerouting decisions. By using the same congestion measure to control both communication behavior and route adaptation, ICDDR creates a closed-loop, self-regulating feedback architecture that treats detection, dissemination, and road-weight-based rerouting as integrated components of a single control mechanism rather than independent processes. The framework is validated through extensive simulations across a Manhattan-style grid and real-world New Delhi urban environments using OMNeT++, Veins, and SUMO under five traffic density conditions. To ensure statistical robustness, results are averaged across five independent random seeds per density level. When evaluated against representative distributed, opportunistic, and centralized traffic-management schemes, ICDDR achieves up to 21
Accurate acquisition of vehicle state information is crucial for active safety control, and state estimation is an important method for obtaining this information. Considering that vehicle states are influenced by the coupling of the dynamic system and road adhesion coefficient, it is necessary to jointly estimate the key states of distributed electric drive vehicles and road adhesion coefficient. However, traditional dynamics-based road adhesion coefficient estimation methods are greatly affected by time lags, which makes it difficult to achieve real-time accurate estimation with Spatiotemporal synchronization, thus affecting the accuracy of vehicle state estimation. This study focuses on distributed drive electric vehicles and proposes a joint estimation method for vehicle states and road adhesion coefficient by integrating visual preview and dynamic estimation. Initially, the road ahead is previewed with a visual sensor, and road segmentation along with type recognition is carried out using the U-Net and MobileNetV2 networks. The identified road type is subsequently used to retrieve the corresponding adhesion coefficient range from a lookup table. Furthermore, a spatio-temporal synchronization strategy and fusion rules for the road adhesion coefficients are developed. Road adhesion coefficient estimates determined by image pre-scanning are integrated into a dynamic estimation method based on low-cost on-board sensors and unscented Kalman filtering algorithm (UKF), which realizes the joint accurate estimation of vehicle state and road adhesion coefficient. Results from the co-simulation experiment using CarSim/Simulink demonstrate that the proposed joint estimation algorithm can effectively and more accurately estimate both road vehicle state.
Drowsiness and distraction by the driver are some of the leading causes of road accidents, and thus there should be a reliable system of real-time monitoring. The conventional techniques that only rely on the Eye Aspect Ratio (EAR) or the visual cues alone are ineffective in recording the dynamics of time and other fatigue features that are not visual. In order to overcome such obstacles, the current paper suggests a new multimodal driver monitoring system that is based on Vision Transformers (ViTs), eye movements based on EAR, and physiological indicators (EEG and heart rate) that are collected via the IoT. The suggested system uses ViTs to extract spatial features with a strong ability, EAR to predict the eye state in real-time, and a CNN-LSTM -fusion model to learn temporal correlations of heterogeneous modalities. The proposed framework contrasts with traditional single-modal approaches by enabling unified multimodal fusion, allowing the model to learn cross-modal relationships for improved robustness. The system is implemented in an edge-cloud system so that it has low-latency real time processing and predictive analytics in the long run. Experimental evaluation on a multimodal dataset collected from 200 participants demonstrates that the proposed approach achieves an accuracy of 97.1
Emergency vehicle coordination in urban traffic represents a safety-critical multi-agent decision problem characterized by partial observability, dynamic constraints, and competing priority demands. Existing rule-based preemption and reinforcement learning approaches are fundamentally limited by reactive policies and the absence of semantic reasoning, restricting their capacity to represent graduated urgency, regulatory context, and inter-agent priority relationships. We propose a Hierarchical Language-Grounded Reinforcement Learning (HLG-RL) framework that integrates LLM-based semantic inference with QMIX-based multi-agent control under the centralized training with decentralized execution paradigm, formalized as a constrained decentralized partially observable Markov decision process (Dec-POMDP). A locally deployed, prompt-engineered LLaMA 3 model processes structured scene serializations to generate context-aware coordination directives. These directives are deterministically translated into action constraints, reward shaping signals, and policy conditioning inputs for agents operations. A deterministic safety projection layer independently enforces physical feasibility and regulatory compliance at every control step, structurally decoupling safety guarantees from upstream semantic variability. Evaluated across SUMO and CARLA under diverse congestion regimes, HLG-RL reduces EV clearance time by 18.5
The development of vehicle-to-everything technology is driving the upgrade of intelligent transportation systems. Collaborative control of heterogeneous vehicles at intersections is essential for improving road network efficiency. This study models connected and manually driven vehicles using an adaptive cruise control model and a two-state safe-speed model, respectively. A hybrid state system and a Gaussian hidden Markov model are introduced to achieve online estimation of driving intentions. Driving safety field theory is integrated into the model predictive control framework, and an improved model predictive control driving risk minimization algorithm is proposed, combined with the branch and bound algorithm. Results show that the improved model predictive control driving risk minimization algorithm reduces the average travel time to 28.6 s at a 90
There is a significant demand for developing the shortest, safest, and most comfortable navigation paths for self-driving vehicles. Meta-heuristic algorithms, inspired by natural processes, are advanced problem-solving techniques designed to explore search spaces and identify near-optimal solutions effectively. However, many nature-inspired algorithms struggle to find the shortest and most appropriate routes. This highlights the need for a robust meta-heuristic algorithm specifically designed for route optimization. In this research, a new meta-heuristic method, Horse Route Evaluation Algorithm (HREA), is proposed. HREA can be applied to a wide range of engineering optimization problems requiring the development of optimal paths. The algorithm demonstrated successful application in self-driving vehicle navigation, delivering superior results. For energy-efficient autonomous vehicle transportation, it is essential to create the shortest navigation path from the starting point to the destination while passing through all required waypoints. Considering safety and comfort, the autonomous vehicle must pass through the most suitable routes with minimum obstructions. To address these challenges, this paper introduces a new Intelligent Path Planning Algorithm (IPPA), which integrates HREA with the optimization principles of Prim’s minimum spanning tree algorithm. The IPPA utilizes a newly designed learning-based fitness function to identify the most suitable path while avoiding obstacles and traffic congestion. Simulations of the Intelligent Path Planning Algorithm (IPPA) demonstrate its efficiency in optimizing autonomous vehicle navigation, showing an 11.11
Urban areas struggle with limited parking spaces due to the rising population within the cities’ fixed boundaries. Private parking spaces can be a better solution if shared efficiently and effectively. However, private parking owners will not share their resources for free and expect minimal overhead during the sharing process, as they are not professionals in this field. In this work, an online auction mechanism for private parking slot allocation (AMPSA) is designed to minimize overhead for private parking owners and encourage citizens to share their empty parking slots at any time. Users seeking parking, known as parkers, place bids to get parking near the desired location. The proposed allocation method schedules the non-overlapping parkers’ requests to available parking lots. Since the parkers may attempt to misrepresent their true value in bidding, the proposed pricing mechanism is designed to promote truthful bidding by offering suitable incentives. Simulation studies on generated datasets with 2,000 parkers and 200 parking owners, evaluated using profit, satisfaction ratio, number of winning parkers, and average waiting time, demonstrate the effectiveness of the proposed approach. AMPSA outperforms one-to-one space allocation by increasing owner profit, parker satisfaction, and the number of winning parkers by around 15
With the advancement of smart city construction, traditional traffic prediction models show obvious deficiencies in coping with unexpected congestion and complex spatio-temporal relationships. The research suggests a dynamic spatio-temporal feature traffic prediction model improved based on the attention mechanism (ASTF-DTPM) to increase prediction accuracy and emergency response. This study enhances the road network topology sensing capability by constructing a dynamic spatial attention module, designing a spatio-temporal fusion framework to realize multidimensional feature synergistic coupling, and innovatively introducing an event-driven feature correction mechanism. The model uses a hierarchical fusion architecture of dynamic graph convolution and multi-scale spatio-temporal attention. This architecture is combined with a two-way gated feature correction algorithm to calibrate prediction errors online. The experimental results indicated that ASTF-DTPM performed well in both short-term and long-term prediction, and the prediction error was reduced by more than 30
Traffic congestion in Vellore has gotten worse, making it harder to go about daily tasks, consuming more fuel, and polluting the environment. This study proposes a Multi-Parameter Dynamic Shortest Route Search Algorithm, which extends the Shortest Route* Search Algorithm introduced by Lakshna et al. (2023) through the incorporation of dynamic multi-parameter edge-cost updating using historical and real-time traffic information.The model combines real-time and historical traffic data, accounting for critical factors such as road types, vehicle characteristics, driver behavior, travel timing, and weather conditions to determine the most efficient travel routes. A weighted graph of Vellore’s road network is constructed, and a heuristic search method is employed to identify the shortest and least-congested routes. Simulation results obtained for the Vellore urban transportation network indicate that the proposed framework reduces average travel time by 17.8
Traffic flows and emergency situations become very difficult to handle by traditional systems as cities become more crowded. The study suggests using both IoT and AI in traffic control to boost the way cities move people. Integrating computer vision and embedded systems into the system allows drivers to be detected instantly and traffic signals to adjust accordingly. By applying YOLOv8 and ESP32, the proposed device uses vehicle counts to change its signals and expect when emergency response is required, so people and machines are safe—without any human supervision. The purpose is to reduce traffic for all drivers and first responders and build smart city infrastructure that can expand as needed. Experiments show that prototypes can handle more traffic, reduce the number of emissions and make cities easier to move through.
Accurate traffic flow forecasting is essential for intelligent transportation systems (ITS). However, existing spatio-temporal forecasting models often suffer from inadequate modeling of dynamic spatial dependencies, insufficient representation of heterogeneous temporal patterns, and excessive computational complexity. To address these issues, this paper proposes a lightweight Dynamic Graph Temporal Decoupling Network (DGTDNet) for multi-step traffic flow forecasting. Specifically, a hybrid graph construction strategy is introduced to combine static road topology with dynamically learned traffic relationships, enabling the model to capture both stable structural dependencies and time-varying spatial interactions. Based on the hybrid graph, a graph convolutional network is employed for spatial feature extraction. In addition, a temporal decoupling module is designed to explicitly separate short-term local fluctuations from long-term temporal dependencies, where a temporal convolution branch captures local dynamics and a Transformer encoder branch models long-range temporal correlations. The two temporal representations are then adaptively integrated through a gated fusion mechanism for final forecasting. Extensive experiments on the public PeMSD4 and PeMSD8 datasets demonstrate the effectiveness of DGTDNet. On the 60-minute forecasting task, DGTDNet achieves Mean Absolute Error (MAE) values of 22.31 and 17.61 on PeMSD4 and PeMSD8, respectively, outperforming MTGNN, the strongest competing baseline, by 3.34
Smart Card Data (SCD) is a valuable source of mobility datasets in public transit systems. While SCD commonly includes the time and location of passengers’ trips, it misses some important trip attributes, such as trip purpose types. Enriching SCD with the trip purpose attribute would extend the design and planning applications of this source of mobility data in public transit systems. In this study, an evolutionary approach, based on concepts of the genetic algorithm, is developed to solve the enrichment problem. The developed approach uses time (start time and duration of the activity, which happens between two subsequent trips) and location (available land use) of the trip destination to infer the trip purpose types. The relation between the time and location of the trips is learned from the Household Travel Survey (HTS). Firstly, the problem is defined and encoded in the genes and chromosomes. Secondly, the objective function is defined and formulated. Thirdly, the selection, cross over, and mutation operators repeatedly evolve each generation of solutions to generate better solutions until the breaking conditions are met. The case study is SCD data from southeast Queensland, Australia. Ultimately, two recent methods in the existing literature are reproduced, and their outcomes are compared with the proposed approach, which shows the high competitiveness between all three methods.
Vehicle trajectory prediction is a key aspect of autonomous driving systems, aiming to infer the future movement of vehicles based on historical trajectories and environmental information. Compared to traditional physics- and rule-based methods, deep learning approaches have demonstrated strong performance in modeling complex spatiotemporal interaction patterns from large-scale datasets and can generate Multimodal Trajectory (MT) hypotheses via probabilistic modeling to represent behavioral uncertainty. This paper systematically reviews the research progress of deep learning in trajectory prediction, focusing on the main thread of "task definition, input representation, interaction modeling, map fusion, multimodal generation, evaluation system, engineering application." First, this paper elaborates on task modeling and output forms; next, this paper summarizes methods such as sequence models, interaction modeling, graph networks, Transformers, and generative models; then, this paper reviews mainstream datasets and evaluation metrics, discussing the importance of closed-loop testing, Probabilistic Calibration (PC) and standardized evaluation protocols for reproducibility; finally, we analyze open issues such as out-of-distribution generalization, long-tail risk, and causal interactions, and highlight trends in foundational models, self-supervised learning, and safe verifiable prediction.
It is essential to anticipate pedestrian motion when planning the trajectories of autonomous vehicles operating in shared environments. Pedestrian future motion patterns must therefore be explicitly considered in any realistic trajectory prediction system. Accordingly, this study presents a systematic review of existing approaches for pedestrian trajectory prediction in the presence of vehicles, with particular emphasis on unstructured environments. In addition to examining how various factors such as prediction uncertainty and behavioral variations are addressed in existing models, this review focuses on pedestrian–vehicle interaction dynamics, as distinct from pedestrian–pedestrian interactions. The review protocol followed the PRISMA guidelines. A comprehensive search across the IEEE Xplore, Scopus, and ACM Digital Library databases identified 1603 peer-reviewed articles, of which 51 most relevant studies were carefully selected that met the defined selection criteria. The review provides a structured summary of datasets deployed in studies involving both vehicle and pedestrian trajectories, and synthesizes key research directions, including the development of large-scale datasets for complex unstructured environments and more expressive representations of interacting agents in deep learning-based prediction frameworks.
