SMART CITY TRAFFIC MANAGEMENT: ACOUSTIC-BASED EMERGENCY VEHICLE DETECTION USING STACKING ENSEMBLE DEEP LEARNING

Abstract

Acoustic data analysis is becoming increasingly vital in smart traffic management systems, particularly for detecting emergency events to enhance traffic safety and efficiency. A key challenge is accurately classifying road noises and emergency vehicle sirens, which is crucial for improving emergency response times and traffic management. This study proposes a Stacking Ensemble Deep Learning model to classify emergency vehicle sirens amidst background noise using traffic data collected via microphone sensors. The model combines Multi-Layer Perceptron (MLP) and Deep Neural Network (DNN) as base learners, with Long Short-Term Memory (LSTM) as a meta-learner. By employing advanced feature engineering techniques including Mel Frequency Cepstral Coefficients (MFCC), spectral centroid, chroma, zero-crossing rate, RMSE, and other acoustic features, the proposed Stacking Ensemble LSTM model achieves superior classification performance. Experimental results on 1,800 audio samples demonstrate that the proposed model achieves 100% classification accuracy, outperforming individual MLP (98%), LSTM (99%), and Stacked LSTM+GRU (99.5%) models. The proposed system demonstrates significant potential for real-time deployment in smart city infrastructure.