Event Detection Using Adaptive Neuro Fuzzy Inference System for a Public Transport Vehicle

Audio surveillance system in a public transport vehicle that detects event like screams and gunshots by classifying signals as normal or in crisis condition using adaptive neuro fuzzy inference system (ANFIS) is presented. Sample audio signals were edited to remove the silent part. Audio signals were divided into frames and represented by its feature. Twelve mel frequency cepstral coefficients (MFCC) represent the feature vector of the frame. Eight audio files were used in the simulation where half of the files represent the normal condition and another half denotes the crisis condition. One hundred data sets from each file were used in training and another 100 data sets from each file were used in validation. The fuzzy inference system was created using the data centers produced using subtractive clustering given the range of influence. Different values of range of influence near the default value of 0.5 were simulated in order to observe the accuracy of the system. The accuracy varies with the range of influence. The best validation accuracy is 87% at range of influence equal to 0.52.