Road Safety Research Articles

Enhancing Deep Learning to Improve Road Safety: An Accident Detection System

This advanced accident detection system significantly improves road safety by enabling timely identification of traffic incidents and rapid emergency response. Current road safety systems often fail to provide accurate and prompt detection of accidents, resulting in delayed emergency services and increased casualties. To address this issue, an advanced system utilizes Convolutional Neural Networks (CNNs) and the You Only Look Once (YOLO) model for real-time accident detection and classification. By analyzing RGB frames and incorporating optical flow data, this system effectively handles dynamic scenarios such as high-speed traffic, varying weather conditions, and complex vehicle movements. Furthermore, a Real-Time Notification feature integrates to automatically alert nearby hospitals and dispatch ambulances immediately upon detecting an accident. Training and testing use the dataset from Kaggle, which includes CCTV footage frames of accidents and non-accidents, ensuring robust performance. This system detects accidents in real- time and promptly sends notifications via email to alert nearby hospitals and dispatch ambulances.

Riding with distraction: Exploring the intention and behaviour of smartphone use while riding among motorcyclists in Vietnam.

Riding with distraction: Exploring the intention and behaviour of smartphone use while riding among motorcyclists in Vietnam.

Automobility violence: the case for adopting tobacco public health policies

ABSTRACT Although tobacco use and road fatalities were recognized as public health issues at roughly the same time, the public health responses have been very different. The accepted wisdom in public health policy is that access to tobacco should be limited, highly taxed, advertising prohibited, visual and textual warnings be mandatory on tobacco products, the obfuscation and lobbying efforts of the tobacco industry publicized. In this paper, we make the case for adopting similar strategies in relation to automobility. As opposed to framing automobility violence as a remediable road safety issue, this paper makes the case for treating automobility violence as an irremediable public health problem. The public health goal with respect to tobacco is not simply to reduce death and disease but to eradicate tobacco use. The public health goal in respect to automobility should be the same. This requires pursuing public health politics oriented towards the dismantling of automobility.

Assessing single motorcycle crash injury severity factor using a novel partial temporally constrained random parameters logit model

Motorcyclists tend to sustain severer injuries in crashes than vehicle occupants. To be able to reduce motorcycle-involved crashes and motorcyclists' injury severities, many previous studies have been conducted to identify injury contributing factors. With the implementation of many countermeasures geared at improving road safety, recent studies have demonstrated temporal variations in crash-contributing factors — implying that countermeasure implementation should not be static over a long period. However, previous studies were not able to easily determine the temporal variations associated with the individual crash-contributing factors. This study adopts the novel partial constraint modelling technique to track the temporal instability of crash factors associated with motorcyclists' injury severity in single-vehicle motorcycle crashes. Using real-time weather information and other influential factors, variables that significantly contribute to motorcyclist injury severity were identified. Weather variables like wind speed, air temperature and visibility were found to significantly affect single motorcycle crash severity.

Multi-body sensor based drowsiness detection using convolutional programmed transfer VGG-16 neural network with automatic driving mode conversion

Many traffic accidents occur nowadays as a result of drivers not paying enough attention or being vigilant. We call this driver sleepiness. This results in numerous unfavourable circumstances that negatively impact people’s life. The identification of driver fatigue and the appropriate handling of such information is the primary objective of this study. Ongoing developments in AI (artificial intelligence) as well as ML (machine learning) within ADAS (Advanced Driver Assistance Systems) have made the application of Internet-of-Things (IoT) technology in driver action recognition necessary. These advancements are dramatically changing the driving experience. This study suggests a novel method for machine learning-based automatic driving change-based drowsiness detection. In this instance, the multi-body sensor detects the driver’s EEG signal and gathers information for brain activity analysis. The wavelet time frequency transform model has been used to examine this signal in order to classify patterns of brain activity. A multi-layer convolutional programmed transfer VGG-16 neural network was then used to classify this examined pattern. This classified signal will cause the automatic driving mode to change. In terms of prediction accuracy, sensitivity, specificity, RMSE, ROC, experimental analysis has been performed for a variety of EEG signal datasets. The goal of this work is to reduce the risks that come with driving while drowsy which will improve road safety and reduce incidents that are related to fatigue.

Security Agents Insight Analysis of the Educational Assessment in Nigeria

An examination is a means of assessing people’s knowledge, competence, or skills in a particular field. It can be through manual or physical contact or a computer-based approach. Achieving the purpose of high-standard examination can only be fulfilled when adequate measures are put in place and steps are taken to implement them. This study presents findings from the baseline study of security agents’ perspectives on conducting and administering examinations in Nigeria. The study adopted a survey approach that was based on the administration of a questionnaire which was carefully formulated to capture the views and opinions of the respondents on the available indices. The administered questionnaire was structured to curate information from the respondents about their knowledge of the stakeholders’ level of infractions and conformity with regulations. The survey attempted to know the extent of deterrent strategies against infractions and corrupt practices, the welfare packages available to security agents, and the examination bodies’ general performances, among others. Twenty thousand and two hundred (20,200) respondents across the six geopolitical zones of Nigeria from different categories of security agencies including the Federal Road Safety Commission (FRSC), Nigerian Civil Defense Corps (NCDC), Police Force, and private security outfits participated in the survey. The statistical-based analysis of data obtained from the survey revealed that the educational system has a competent workforce with a significant number of tertiary-educated individuals, a significant level of compliance of stakeholders to examination regulations, and effective enforcement of rules by security agencies. The analysis also revealed a substantial level of inadequacies with the operating environment and mechanisms for welfare and mobility packages as well as gaps in certain aspects of the examination process which may be connected to inefficient communication, logistical challenges and inadequate resources. The study justified the need for more attention to improving on the deterrent measures against examination misconduct through increased vigilance and addressing some factors that militate against the existing measures for curbing corrupt practices during the examination.

A randomized, placebo-controlled, double-blind, pilot study of cannabis-related driving impairment assessed by driving simulator and self-report.

In the context of increasing cannabis use, understanding how cannabis affects specific driving behaviors is crucial in mitigating risks and ensuring road safety. The current study included 38 adults aged 18-40 years, administered a single 0.5 g acute dose of vaporized cannabis (5.9% Tetrahydrocannabinol (THC), 13% THC or placebo) in a randomized, within-subject, double-blind, counterbalanced design. Throughout each of the three, 8-h assessment days, at 4 time points, participants underwent simulated driving tests, including lane-keeping, car following, and overtaking tasks, capturing 19 behavioral metrics. An SPSS linear mixed model assessed the main effects of dose, time, and dose × time. During lane-keeping, participants exhibited reduced steering reversal rates up to 5.5 h following 13% THC and 3.5 h for 5.9%. For car following, participants showed reduced pedal peak-to-peak deviation and reversal rates, persisting for 1-3 h post-dose (only at 13% THC). During overtaking, following 13% THC, subjects demonstrated a shorter median gap to passed cars, lower time-to-potential collision, and more time in the oncoming lane. Drug effects on driving metrics improved gradually, to varying degrees over time. Approximately 66% of participants reported willingness to drive, despite subjective awareness of being impaired and objectively worse driving performance. Our study reveals for the first time long-lasting cannabis-induced impairments across multiple driving behaviors, that extend beyond the typical 3-h window explored in most previous research. The observed discrepancy between participants' willingness to drive and their actual impairment highlights an important public safety concern. In addition, the lack of correlation between cannabinoid metabolite concentrations and driving performance challenges the reliability of blood THC levels as impairment indicators, emphasizing the need for a multifaceted approach to assessing cannabis-impaired driving risk.

Knowledge, Perceptions, and Reactions of Lagos Road Users to Road Safety Messaging on Traffic Radio: An Analytical Study

Road traffic safety remains a pressing challenge in Lagos, Nigeria, where urbanisation, population growth, and vehicular congestion contribute to frequent road accidents and fatalities. This study explores the knowledge, perceptions, and reactions of Lagos road users to road safety messaging disseminated through Traffic Radio, a medium specifically dedicated to traffic updates and safety advocacy. Employing a mixed-methods research design, the study collected data from 500 respondents using structured surveys, complemented by focus group discussions with commercial drivers, private vehicle owners, motorcyclists, and pedestrians. Findings reveal that while awareness of road safety campaigns on Traffic Radio is high (79% of respondents), there are significant gaps in translating this awareness into sustained behavioural changes. Socio-economic factors, such as income levels and educational background, alongside infrastructural challenges, were identified as key barriers to compliance with road safety measures. Moreover, the study highlights variations in perception, with younger road users and commercial drivers showing lower levels of engagement with Traffic Radio messaging compared to private vehicle owners and older demographics. Qualitative insights suggest that the timing, content, and tone of safety messages significantly influence audience receptiveness and the perceived relevance of the campaigns. Recommendations include the development of more targeted, culturally sensitive campaigns, the integration of multilingual content to reach diverse audiences, and increased partnerships between Traffic Radio, government agencies, and community organisations to enhance public education on road safety. This study contributes to the body of knowledge on the role of media in fostering behavioural change, offering evidence-based insights for policymakers, media practitioners, and urban planners in addressing road safety challenges in high-density urban areas.

Single-level Discrete Two Dimensional Wavelet Transform Based Multiscale Deep Learning Framework for Two-Wheeler Helmet Detection

INTRODUCTION: A robust method is proposed in this paper to detect helmet usage in two-wheeler riders to enhance road safety. OBJECTIVES: This involves a custom made dataset that contains 1000 images captured under diverse real-world scenarios, including variations in helmet size, colour, and lighting conditions. This dataset has two classes namely with helmet and without helmet. METHODS: The proposed helmet classification approach utilizes the Multi-Scale Deep Convolutional Neural Network (CNN) framework cascaded with Long Short-Term Memory (LSTM) network. Initially the Multi-Scale Deep CNN extracts modes by applying Single-level Discrete 2D Wavelet Transform (dwt2) to decompose the original images. In particular, four different modes are used for segmenting a single image namely approximation, horizontal detail, vertical detail and diagonal detail. After feeding the segmented images into a Multi-Scale Deep CNN model, it is cascaded with an LSTM network. RESULTS: The proposed model achieved accuracies of 99.20% and 95.99% using both 5-Fold Cross-Validation (CV) and Hold-out CV methods, respectively. CONCLUSION: This result was better than the CNN-LSTM, dwt2-LSTM and a tailor made CNN model.

Assessment of safety orientation in driving skills aligned with performance: a data-triangulation approach

Accurate assessment of Subjective Driving Skills (SDS) is crucial for improving road safety, as direct methods are often biased and do not align well with actual driving performance. This study aimed to develop an unbiased SDS assessment method aligned with driving performance. The specific objectives are (1) reducing bias in SDS assessments, (2) verifying alignment between assessed safety orientation and ground driving performance, and (3) exploring the influence of socio-demographic factors on safety orientation. A combined questionnaire and photographic speed survey were conducted among 389 experienced car drivers in Mangalore, India. Factor analysis, a Double Lane Change (DLC) test conducted on the ground with a test vehicle equipped with Inertial Measurement Unit (IMU) sensors, correlation analysis and multiple linear regression were performed. Factor analysis confirmed the two-factor structure: Perceptual-Motor (PM) and safety skills. Further, PM and safety skills scores were calculated using factor loadings, and safety orientation was determined from their difference. DLC results showed that the assessed safety orientation aligned with the ground performance. Correlation and regression analyses showed that male drivers perceived slightly higher PM skills than female drivers. PM skills decreased with age, while safety orientation increased. Academic education had no significant effect on safety skills or safety orientation. While on-road exposure improved PM skills, weekly driving distance reduced safety orientation. Formally trained drivers had slightly higher safety skills and safety orientation than lay-instructed drivers. These findings provide several valuable insights for enhancing road safety. They suggest that safety programs address overconfidence in male drivers, incorporate road safety awareness into educational curriculums, and offer enhanced training for all experienced drivers. Younger drivers may benefit from targeted safety campaigns, while professional drivers could require specialised safety programs. Regular safety assessments and refresher courses are crucial for maintaining safety awareness, particularly for drivers with higher weekly driving distances.

Lane Detection for Self-Driving Cars

he project focuses on developing a real-time vehicle detection and lane analysis system using computer vision techniques. The system is designed to process video inputs, detect vehicles, analyze lane curvature, and provide safety suggestions to drivers. The application leverages YOLO (You Only Look Once) for vehicle detection, optical flow for speed estimation, and Hough Transform for lane detection. The system is implemented as a Flask-based web application, allowing users to upload video files, process them, and view the results with overlays indicating vehicle distances, lane status, and safety suggestions. The project aims to enhance road safety by providing real-time feedback to drivers about their surroundings. Keywords: Real-time Vehicle Detection, Lane Detection, YOLO, Optical Flow, Hough Transform, Road Safety, Computer Vision, Video Processing, Lane Curvature Analysis, ADAS

On-Road Vehicle Breakdown Assistant

The On-Road Vehicle breakdown Assistant is a web-based system developed using PHP to provide real-time assistance to drivers and vehicle owners. This system offers features such as vehicle diagnostics, route optimization, emergency alerts, and maintenance reminders. By integrating PHP with MySQL for data storage and JavaScript for an interactive interface, the system ensures a seamless user experience. The assistant helps in monitoring vehicle health, notifying users of potential issues, and suggesting nearby service stations based on GPS data. Additionally, it enhances safety by offering SOS services in case of emergencies. The system's intuitive dashboard allows users to track fuel efficiency, trip history, and servicing schedules. This PHP-based solution is designed to improve vehicle performance, minimize breakdowns, and optimize travel routes, ultimately enhancing road safety and driving convenience. Future improvements may include AI-driven predictive maintenance and voice-based assistance for a more interactive experience.

Harnessing Smartphone Sensors for Enhanced Road Safety: A Comprehensive Dataset and Review

Severe collisions can result from aggressive driving and poor road conditions, emphasizing the need for effective monitoring to ensure safety. Smartphones, with their array of built-in sensors, offer a practical and affordable solution for road-sensing. However, the lack of reliable, standardized datasets has hindered progress in assessing road conditions and driving patterns. This study addresses this gap by introducing a comprehensive dataset derived from smartphone sensors, which surpasses existing datasets by incorporating a diverse range of sensors including accelerometer, gyroscope, magnetometer, GPS, gravity, orientation, and uncalibrated sensors. These sensors capture extensive parameters such as acceleration force, gravitation, rotation rate, magnetic field strength, and vehicle speed, providing a detailed understanding of road conditions and driving behaviors. The dataset is designed to enhance road safety, infrastructure maintenance, traffic management, and urban planning. By making this dataset available to the community, the study aims to foster collaboration, inspire further research, and facilitate the development of innovative solutions in intelligent transportation systems.

Risky scooter riders pay more: a usage-based insurance study of a scooter risk-aware fingerprint model without privacy issues

Usage-based insurance (UBI) has emerged as a promising strategy in the auto insurance industry by tailoring premiums to individual driving behaviors and associated risks. However, privacy concerns regarding the collection and storage of GPS data have limited its adoption. Moreover, current UBI research has largely overlooked scooters, whose operational dynamics differ markedly from cars. To address these limitations, this study proposes a UBI framework tailored for scooters. The framework introduces a multi-dimensional risk evaluation and a risk-aware fingerprint (RAF) model, developed using deep learning and machine learning techniques, which specifically estimates driving risk without compromising privacy. In Experiment 1, the RAF model achieved macro-F1 scores of 78.7%, 80%, and 80.9% at different thresholds, while Experiment 2 yielded scores of 89.6%, 87.1%, and 80.9%. Although the RAF model did not perform best over entire journeys, it reliably identified high-risk riders. These results suggest that the RAF model can effectively replace traditional models based on raw behavior features, thereby enabling the implementation of variable insurance premiums. Such premiums may incentivize safer driving practices among high-risk riders and indirectly enhance road safety. Future research may extend this framework to food delivery services through collaborations with insurance companies, providing scooter drivers with traffic conflict-related indemnity.

Optimized Real-Time Decision Making with EfficientNet in Digital Twin-Based Vehicular Networks

Real-time decision-making is vital in vehicular ad hoc networks (VANETs). It is essential to improve road safety and ensure traffic efficiency and flow. Integrating digital twins within VANET (DT-VANET) creates virtual replicas of physical vehicles, allowing in-depth analysis and effective decision-making. Many vehicular ad hoc network applications now use convolutional neural networks (CNNs). However, the growing model size and latency make implementing them in real-time systems challenging, and most previous studies focusing on using CNNs still face significant challenges. Some effective models with sustainable performances have recently been proposed. One of the most advanced models among them is EfficientNet. One may consider it a family of network models with significantly fewer parameters and computational costs. This paper proposes EfficientNet-based optimized real-time decision-making in the DT-VANET architecture. This paper investigates the performance of EfficientNet in digital-based vehicular ad hoc networks. Extensive experiments have proved that EfficientNet outperforms CNN models (ResNet50, VGG16) in accuracy, latency, computational efficiency, and convergence time, which proves its effectiveness in real-time applications of DT-VANET.

Real-Time Traffic Flow Prediction Model Based on Deep Learning: Taking Smart Highways as an Example

Architectural designers are adopting changes in the sector of traffic management. Advancements and innovations mean a lot in designing predictive models. This is practical in urban areas with high traffic flow. Applying the scenario of smart highways, we develop a model based on deep learning techniques to address challenges faced in this sector. We further our study by utilizing Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs) to capture temporal dependencies and spatial correlations within traffic data. A comprehensive dataset, including real-time sensor data from intelligent highways, was used to train and validate the model’s performance. The results of this study show that the deep learning model significantly outperforms traditional methods. This provides superior accuracy and faster response times when predicting traffic conditions. Also, the model demonstrates robustness in adapting to fluctuating traffic patterns, like congestion, accidents, and peak-hour traffic. This makes the model to be highly reliable for dynamic traffic management. By enabling efficient traffic flow predictions, the model holds promise for reducing congestion, enhancing road safety, and optimizing the overall performance of intelligent highway systems. Thus, this paper provides a strong foundation for the implementation of intelligent transportation systems by highlighting the transformative potential of deep learning in future traffic management strategies.

Nonparametric Comparative Analysis of Driver Behaviors in Signalized and Non-Signalized Roundabouts: A Study on Road Safety in Qatar

This study investigated and compared driver behaviors at signalized and non-signalized roundabouts in Qatar, focusing on turn signal usage, lane change behavior, and correct lane usage. The primary objectives were to determine the frequency of turn signal usage, assess correct lane usage, analyze lane change behavior, and compare these behaviors between the two types of roundabouts. Data were collected through a field study at selected roundabouts, where driver behaviors were observed and analyzed. The results revealed significant differences between signalized and non-signalized roundabouts. Turn signal compliance was higher in signalized roundabouts (up to 45%) compared to non-signalized roundabouts (20%). The rate of lane change in signalized roundabouts was observed to be 31%, whereas it was 14% in non-signalized roundabouts, and correct lane usage compliance was higher in signalized roundabouts (60%) compared to non-signalized roundabouts (35%). These findings suggest that traffic signals contribute to safer and more predictable driver behavior, although congestion and long waiting times in signalized roundabouts present challenges. The study recommends improving signage visibility, optimizing signal timings, enhancing road markings, and enforcing traffic regulations to address these issues. The findings can inform traffic engineers and policymakers in enhancing the safety and efficiency of roundabouts.

Improving Road Safety of Bicycles through Vibration Based Energy Generation

The world is currently prioritizing sustainable energy concepts, recognizing the significance of electric power in modern living and technological advancement. Amidst a recent fuel crisis in Sri Lanka, bicycles have gained popularity as transportation vehicles and globally renowned sporting equipment, contributing significantly to eco-friendly mobility in urban and suburban areas. In this research context, it focuses on harnessing piezoelectric energy from bicycle vibrations to enhance eco-friendly mobility and bicycle safety in Sri Lanka. It addresses as a solution to environmental concerns related to disposable batteries, promoting the use of rechargeable batteries, and offering a sustainable alternative to traditional bicycle lighting systems.

System Dynamics in Road Safety: A Comprehensive Overview with Selected Use-Cases

Road safety is a critical concern worldwide, with traffic accidents causing significant loss of life, injuries, and economic costs. Traditional road safety approaches often focus on isolated interventions without adequately considering the complex interactions within the transportation system. System Dynamics (SD) offers a comprehensive framework for understanding and improving road safety by modeling the intricate relationships and feedback loops that influence system behavior over time. This paper provides a comprehensive overview of the application of System Dynamics in road safety, complemented by selected simple use cases to illustrate its effectiveness.

Last Mile Urban Freight Distribution: A Modelling Framework to Estimate E-Cargo Bike Freight Attraction Demand Share

Urban freight transportation is facing significant challenges due to increasing demand, driven by globalization, e-commerce growth, and the adoption of just-in-time logistics. These trends have led to rising vehicle flows in urban areas, negatively impacting sustainability, economic efficiency, and road safety. In response, cities are exploring innovative last-mile delivery strategies that emphasize sustainability, flexibility, and cost efficiency. Among these strategies, cargo bikes—particularly electric cargo bikes (e-cargo bikes)—are emerging as promising low-emission solutions for urban freight distribution. However, despite their potential, a generalized methodology for estimating their demand share in urban contexts remains underdeveloped. This study proposes a comprehensive modelling framework to evaluate the freight demand share that can be addressed by e-cargo bikes, integrating quantity, restocking service, modal, and delivery sub-models, calibrated using data from a case study in Italy. The results demonstrate that e-cargo bikes could fulfil up to 20% of urban freight demand, depending on the category of goods transported, and underscore the feasibility of integrating e-cargo bikes into urban logistics systems. However, critical challenges related to scalability and cost-effectiveness persist, highlighting the need for further research and reliable cost data to support broader implementation.