Moving object classifier based on UWB radar signal

Abstract

A novel moving object classification system using UWB radar and classifier based on decision tree structure are proposed. By using the proposed radar system, we construct UWB radar signal database by considering two movements and four moving directions of human and dog. The proposed classifier is based on nonlinear support vector machine (SVM) using RBF kernel and use linear predictive code (LPC) coefficients as feature vector. By evaluating performance of the proposed decision tree structures, we obtain the best classification results when the first level SVM classifies type of movement and then the second level SVM classifies moving object. The correct classification probability ranges from 93% up to 97%. The proposed system and classifier can be used for efficient human and dog classification and can be applied to other moving objects classification as well.