Vehicle Accident Detection and Notification System

Abstract

As it is observed these days, the roads are crowded with different kinds of vehicles, and as a result, accidents have become more dangerous every day. In this paper, a vehicle accident detection and notification system is designed by using road surveillance camera data and machine learning techniques to use the recorded footage from the surveillance cameras to detect the accident status, activate an alerting sound, and send the notification message as a Google email (Gmail) to the specialist. The system model was made by combining two algorithms, convolutional neural networks (CNN) and support vector machines (SVM), to make a CNN-SVM hybrid model, which was then trained using two datasets. The first dataset contains 4814 images with sizes (28, 28) and extensions (jpg), and the second dataset contains 990 images with sizes (32, 32) and extensions (jpg). The outcomes scores of the evaluation matrices are: accuracy for the first dataset is 99.74% and loss is 1.14%, and for the second dataset, accuracy is 98.88% and loss is 3.39%. When testing it with real-world data, it achieved its objectives in 30 seconds, with accuracy reaching 100%.