Development of a Deep Learning-Based System for Road Traffic Anomaly Detection

Abstract

Continuous supervision is required to monitor real-world traffic surveillance recordings and take necessary steps in the event of tragic incidents. Nevertheless, constantly monitoring them under human supervision is laborious and prone to errors. Hence, a deep learning technique has been suggested to automatically identify and pinpoint road accidents by framing the issue as anomaly detection. This study focuses on developing an advanced system that uses deep learning approaches to detect irregularities in road traffic. Our research aims to meet this necessity by utilizing the potential of deep learning techniques that are specifically adapted to the complexities of road traffic monitoring. To overcome the limitations of lack of knowledge of the causes of road traffic, the system employs Super-Resolution Generative Adversarial (SRGAN) and Convolutional Neural Network (CNN) models to classify the images obtained from the dataset. The study performed excellently well using CNN and SRGAN with an accuracy of 100%, precision of 100%, recall of 100%, f1-score of 100% and ROC value of 100%.