Omar Alshamsi

Abstract

Road traffic crashes cause over 1.19 million fatalities per year worldwide and represent a major challenge for urban safety (World Health Organization, 2023; World Bank Group & WRI India, 2020). Smart city initiatives aim to use technology to reduce such losses through rapid accident detection and response (Arefin et al., 2025). This proposal outlines a vision for a Road Accident Detection and Response System that employs computer vision and deep learning to identify accidents from video feeds and immediately notify emergency services. The core of the system will be a custom-trained YOLOv8 object detection model, leveraging a dataset of annotated accident images (cars, buses, trucks, motorcycles) from Roboflow Universe (Geetha et al., 2024). Exploratory Data Analysis (EDA) will characterize the dataset (class distribution, bounding box sizes, environmental conditions) to guide training. The model will be developed locally on an NVIDIA RTX 2060 GPU using the Ultralytics YOLOv8 framework (Ultralytics, 2023). A data.yaml file will define the custom accident classes and training/validation splits, following YOLO conventions (Ultralytics, 2023). After training, the system’s performance will be rigorously evaluated on held-out data by computing precision, recall, F1-score, and mean Average Precision (mAP) at various IoU thresholds (Geetha et al., 2024). These metrics will quantify detection accuracy (for instance, YOLOv3 achieved approximately 57.9% mAP@0.5 in 22 milliseconds of inference time) (Redmon and Farhadi, 2018), confirming real-time viability. The final system will integrate the detection model with an alert mechanism to dispatch notifications (e.g., SMS, email) to first responders or traffic management centers (Arefin et al., 2025). Key deliverables include the trained model, a codebase for real-time accident detection, a detailed analysis of dataset and model performance, and a documented prototype implementation. This work fills a gap in smart-city safety by applying the latest YOLOv8 technology to automated accident detection and rapid notification (Khalili & Smyth, 2024). It builds on recent research showing YOLOv8’s superior accuracy and speed (Ultralytics, 2023; Ghazzaoui and Kubra, 2025), but extends it to the critical domain of traffic incident response, aiming to reduce emergency response times and save lives. Recent work also reports strong accident-detection results with an improved YOLO11 (‘YOLO11-AMF’) that adds a linear-attention module, an asymptotic FPN, and a Focaler-IoU loss, improving precision/recall and mAP over the YOLO11n baseline on a curated accident dataset (Li, Huang & Lai, 2025)

Library of Congress Subject Headings

Traffic safety--Automation; Computer vision; Optical pattern recognition; Deep learning (Machine learning); Smart cities

Publication Date

9-10-2025

Document Type

Thesis

Student Type

Graduate

Degree Name

Professional Studies (MS)

Department, Program, or Center

Graduate Programs & Research

Advisor

Sanjay Modak

Advisor/Committee Member

Parthasarathi Gopal

Campus

RIT Dubai

Plan Codes

PROFST-MS

Download

Share

COinS