A Novel Smart Driver Assistance System to Reduce Traffic Congestion Using Metaheuristic Algorithms and Deep Learning Methods

Abstract

Cities across the globe have faced serious challenges of road accidents and traffic jams in recent days thereby increasing huge human and economic losses. The key to improving Advanced Driver Assistance Systems (ADAS) is precise analysis and modeling of driver behavior in complex driving scenarios. To deal with such challenges, acoustic-based ADAS systems have come into the limelight due to their cost-effectiveness and versatility. In this study, we suggest an AI-based Intelligent transportation system with smart ADAS technology to improve road safety and reduce traffic jams using acoustic data analysis techniques. We enhanced our integrated ITS and ADAS system using three acoustic datasets: IDMT Traffic for traffic assessment, LSA Vehicles for emergency sirens and road noise analysis, and RQA Road for road quality evaluation. The proposed methodology includes signal preprocessing by eliminating undesirable artifacts from the sound data, performed with a modified coati optimization (MCO) algorithm. Afterward, a feature extraction phase employs the Y-Net pretrained architecture that reveals hidden features that were optimized by the Emperor Penguin Optimization (EPO) algorithm to address dimensionality issues. Finally, a Recurrent Artificial Neural Network (RANN) with hybrid recurrent features is used to detect abnormalities in roadways and provide recommendations for drivers. The validation results on three benchmark datasets show promising results for prediction accuracy and false positive rate. The global results conclude that the proposed solution realized a superior classification accuracy score (96.252 %), precision (0.95), recall (0.94), and F-scores (0.95) compared to various baseline methods such as SVM, XG-Boost, t-SNE, CNN+LSTM, and SVM+LSTM.