Towards Zero Road Fatalities: Automated Road Accident Detection Using Deep Learning and Transfer Learning with Synthetic Data

Abstract

Over a million people have died in road accidents every year, making them one of the world's top sources of injury and death. Human error and delayed incident response are major contributors to road deaths, even with improvements in vehicle design and road infrastructure. The need for precise, real-time solutions to quickly identify traffic incidents is growing as the transportation industry moves toward automation and intelligent mobility. Achieving zero road deaths in the age of intelligent transportation is still a global priority. This study proposes a robust deep-learning framework for automated road accident detection by harnessing the synergy of transfer learning and synthetically generated data. Addressing the limitations of scarce annotated accident imagery, we synthesize a diverse dataset representing normal and alarm (accident) conditions to train and evaluate state-of-the-art convolutional neural network models. Leveraging ImageNet-pretrained architectures, including MobileNetV2 and EfficientNetB1, we implement fine-tuning strategies that preserve generalizable visual features while adapting classification layers for binary decision-making. Experimental results demonstrate that EfficientNetB1 achieves superior detection performance with a 95% accuracy, 0.88 F1 score, and 0.90 MCC, outperforming MobileNetV2, which recorded a 93% test accuracy and 0.81 F1 score. Our methodology further explores and compares a suite of advanced CNN models, Xception, VGG16/19, ResNet50/V2, InceptionV3, and EfficientNetB7, offering insights into architecture selection for real-time, resource-constrained deployment.