Traffic Safety Measures and Assessment

Abstract

Traffic safety is undergoing a profound transformation, driven by advances in data science, sensing technologies, and computational modeling. Proactive approaches are enabling the early identification of potential hazards, real-time decision-making, and the development of smarter, safer transportation systems. This Special Issue summarizes recent progress in traffic safety assessment, highlighting the application of emerging tools such as machine learning, explainable artificial intelligence, and computer vision. These innovations are used to predict crash risks, evaluate surrogate safety measures, and automate the analysis of behavioral data, contributing to more inclusive and adaptive safety frameworks, particularly for vulnerable road users such as pedestrians and cyclists. The research also addresses key challenges, including data integration across diverse sources, aligning safety metrics with human perception, and ensuring the scalability of models in complex environments. By advancing both technical methodologies and human-centered evaluation, these developments signal a shift toward more intelligent, transparent, and equitable approaches to traffic safety assessment and policy-making.

1. Introduction

In recent years, traffic safety has undergone a significant transformation, driven largely by rapid advances in data science, sensing technologies, and computational modeling. Traditional methods of evaluating road safety, relying on historical crash data, are increasingly being supplemented or replaced by proactive approaches. These innovations enable researchers and practitioners to identify potential hazards before crashes occur, support real-time decision-making, and design smarter, safer transportation systems [1].

A major catalyst in this shift has been the integration of technologies such as machine learning (ML), artificial intelligence (AI), digital twin modeling, and connected vehicle (CV) systems [2,3,4,5]. These technologies allow for the collection and analysis of high-resolution data related to vehicle dynamics, road environments, driver behavior, and infrastructure conditions. For example, AI-based algorithms can predict crash likelihood based on trajectory patterns, weather data, and traffic signal timing [6]. Similarly, digital twins offer real-time simulation environments that can be used to evaluate the safety performance of intersections, road segments, and transportation networks under varying conditions [7]. Deep learning methods are applied to develop traffic crash models capable of predicting both crash occurrence and severity [8,9]. Spatio-temporal methods can characterize accident progression patterns and distributions [10,11]. In addition, machine learning algorithms can be used to model driving behavior, and their integration into driver-assistance systems enables the development of proactive strategies for enhancing road safety [12,13,14].

Despite these advances, several key challenges persist. First, while data-driven methods have proven promising in controlled environments, their scalability and robustness in real-world applications remain areas of active research. The fusion of heterogeneous data sources such as vehicle sensors, road infrastructure, and user-reported feedback presents technical and methodological hurdles, particularly regarding data privacy, standardization, and reliability [15].

Second, surrogate safety measures (SSMs), such as time-to-collision and post-encroachment time, have emerged as valuable tools for assessing near-miss incidents and identifying hazardous interactions [16]. However, there is still a lack of consensus on the most effective indicators, how they should be calibrated across different traffic contexts, and how they correlate with actual crash outcomes [17]. Furthermore, many existing models still overlook key variables such as human cognitive behavior, environmental disruptions, and the complex interplay between autonomous and human-driven vehicles [18].

Third, the increasing presence of vulnerable road users (VRUs), including pedestrians, cyclists, and micromobility users, demands more inclusive safety frameworks [19]. Existing traffic models and safety assessments are often vehicle-centric and may fall short in capturing the unique risks and behaviors associated with non-motorized users. This limitation underscores the need for more adaptable, multi-modal safety evaluation methods that reflect the diversity of urban mobility [20].

In response to these evolving needs, this Special Issue brings together papers that reflect the latest developments in traffic safety technologies and methodologies. These works collectively explore the application of intelligent systems, behavioral analysis, and multimodal safety frameworks to improve predictive accuracy and decision support.

As transportation systems grow in complexity, the integration of intelligent, adaptable, and inclusive safety tools will be essential. This Special Issue captures the state of the art while also identifying pathways for future research and innovation, particularly in the direction of AI deployment, behavioral integration, and equitable safety design.

2. Overview of Published Articles

This Special Issue features papers that advance the application of data-driven, intelligent, and human-centered approaches in traffic safety assessment. The studies span a wide range of topics, including crash severity modeling (List of Contributions, 1–4), safety evaluation (List of Contributions, 5, 6), tunnel lighting simulation (List of Contributions, 7), heavy vehicle crash evaluation (List of Contributions, 8, 9), driving behavior (List of Contributions, 10, 11), and the risk profiling of vulnerable road users (List of Contributions, 12). Methodologically, they integrate tools such as machine learning, explainable artificial intelligence, computer vision, simulation modeling, fuzzy logic, and ensemble classifiers. Across these works, key contributions include improved predictive accuracy for crash outcomes, enhanced interpretability of risk factors, the alignment of surrogate indicators with real-world driver perception, and the automation of safety analysis using visual and behavioral data. Together, these papers not only demonstrate the maturity of emerging safety technologies but also provide actionable insights to support safer infrastructure design, targeted interventions, and more inclusive traffic safety strategies. Below, we highlight three representative contributions to illustrate the diversity and impact of the research presented.

One study applies Explainable Artificial Intelligence (XAI) to investigate the connection between aggressive driving behaviors and traffic accidents involving elderly pedestrians [21]. Using crash records and vehicular driving data, the authors train an XGBoost model and apply SHAP (Shapley Additive Explanations) to interpret the influence of variables such as rapid acceleration, deceleration, and speeding. The results show that these aggressive behaviors, combined with spatial factors like elderly population density and infrastructure layout, significantly increase crash risk. This research demonstrates how XAI not only improves predictive accuracy but also enhances the transparency and policy relevance of machine learning models for safety planning.

Another article explores how surrogate safety indicators align with actual human perceptions of danger during traffic events in the driving context [22]. The authors use controlled video experiments and participant feedback to compare event severity ratings with traditional indicators such as time-to-collision and post-encroachment time. The study reveals a mismatch in certain cases, suggesting that surrogate measures may not fully capture perceived risk from a human-centered perspective. This research provides a critical lens on the applicability of widely used metrics and encourages the development of more behaviorally grounded assessment frameworks.

In the domain of image-based analysis, one study evaluates the performance of three popular deep learning models, AlexNet, ResNet-50, and VGG-19, for classifying features in pedestrian crash diagrams [23]. These visual records, often used in police reports, contain valuable information that can aid in crash reconstruction and hotspot detection. The study benchmarks the models in terms of accuracy, computational efficiency, and feature extraction quality. This contribution is significant in pushing forward the automation of visual crash data processing for large-scale safety evaluations.

Collectively, the articles in this Special Issue demonstrate the potential of integrating machine learning, human-centered evaluation, and computer vision to support more intelligent and inclusive traffic safety management. By improving model transparency, aligning safety metrics with human behavior, and undertaking large-scale analysis, these studies offer valuable tools and perspectives for researchers, practitioners, and policymakers.

3. Conclusions

This Special Issue brings together a diverse collection of research that advances the technological frontiers of traffic safety assessment. By leveraging techniques such as machine learning, deep learning, surrogate safety analysis, simulation modeling, and explainable AI, the featured research demonstrates a clear movement toward proactive, predictive, and context-aware safety frameworks. The articles provide practical methods for identifying high-risk conditions, improving crash analysis accuracy, and enhancing the interpretability of safety models for policy and planning. Importantly, the Special Issue also expands the discussion to include human perception, vulnerable road users, and automated analysis of visual safety data. Looking ahead, future research should aim to validate these methods in diverse real-world environments, advance multi-modal safety modeling, and strengthen the integration of intelligent systems with human factors to support safer and more adaptive transportation networks.