Fast Classification for Identification of Vehicles on the Road from Audio Data of Pedestrian’s Mobile Phone

Vehicle classification using machine learning is introduced in numerous application areas such as traffic management in foggy and rainy weather, vehicle recognition in driver-free vehicles, for pedestrians to give alarms, also sent a notifications to phone handling persons while walking, etc. Numerous research papers are available with ≥ 90% accuracy, but only a few of them are available for real time classification using 1-D audio signals. The main idea of this work is to classify various vehicles within a short span of time with acceptable accuracy. The real-time data collection is done at three different locations using various mobile phones. The Mel frequency cpectrum coefficient (MFCC) is used to extract features from audio samples to train and test classifiers. Machine learning classifiers such as K-nearest neighbor (KNN), Support Vector Machine (SVM), and LightGBM are used to classify a number of vehicles. The experimental results we presented for comparison of prediction timing and accuracy aimed for real-time implementation.