Emergency Vehicle Direction Detection using Mel-Frequency Cepstral Coefficients and Deep Learning

Emergency vehicles face difficulty in navigating through traffic and locating the shortest route to the incident site. In the developing countries or under disaster conditions when limited communication and Intelligent Transportation System (ITS) infrastructure is available, the situation becomes more challenging. This paper addresses these challenges by leveraging the unique sound generated by emergency vehicles to detect the direction of the emergency vehicles. We use the Mel-frequency cepstral coefficients (MFCC) and a Convolutional Neural Network (CNN) stacked with Long Short-Term Memory (LSTM) layers to accurately detect the movement direction of outgoing emergency vehicles at a junction. MFCC facilitates effective feature extraction from audio signals, while the CNN-LSTM architecture enables robust temporal and spatial pattern recognition. The information can be used to adapt the traffic lights at the downstream junction to enable the uninterrupted movement of the emergency vehicle through the junction. The experiment results demonstrate 98.56% accuracy in detecting movement direction of an emergency vehicle. The solution can improve emergency response under resource constrained environment and in turn improve public safety and well-being.