Ultrasonic sensor-based human detector using one-class classifiers

Abstract

Human detection is vital to many applications, for example, human-robot interaction, unattended ground sensor systems, smart rooms, etc. In this paper we investigate the application of a one-class classifier to the problem of human detection using solely ultrasonic sensors. Our approach is based on fuzzy rules that are extracted from the signal features in time and frequency domains. The performance of the human detector system (classifier) is assessed in terms of accuracy, true positive, and false positive rates by conducting several experiments. The results show that the system attains high accuracy and high recall with a very low false alarm rate. We have also compared its performance with the widely used support vector machine (SVM) classifier and found that our system is relatively better than the one-class SVM.