Intelligent City Traffic Management: Acoustic-Based Vehicle Detection with Stacking-Based Ensemble Deep Learning

Abstract

In the realm of smart city traffic management, improving urban mobility requires effective monitoring and prompt emergency reaction. The current stacking-based ensemble deep learning method for acoustic-based vehicle recognition is extended in this study with an emphasis on enhancing classification accuracy for road noise and emergency vehicle sirens, such those of ambulances. Although Fully Connected MLP, Deep Neural Network (DNN), and LSTM networks were stacked in the original method, new research improves performance even more by adding a lightweight Gated Recurrent Unit (GRU) layer. The LSTM classifier is combined with GRU, which is well-known for its capacity to increase processing speed and prediction accuracy by streamlining the architecture of recurrent neural networks, to maximize handling of big datasets and increase accuracy even further. The expanded model's LSTM and GRU combination outperforms earlier setups, achieving an astounding 100% accuracy. This model more accurately identifies road noise and emergency vehicle noises by utilizing acoustic characteristics as fundamental frequency, loudness, and amplitude. The suggested method provides a novel approach to real-time traffic management and emergency vehicle prioritizing in smart cities, outperforming conventional algorithms that only use Mel Frequency Cepstral Coefficients (MFCC) or Mel spectrograms. Accuracy, precision, recall, and Fscore measures are used to assess the performance of all models, including the extended LSTM-GRU method, showing how effective the suggested system is for managing traffic in cities in the future