Detection and Localization of People Inside Vehicle Using Impulse Radio Ultra-Wideband Radar Sensor

Abstract

Nowadays, various sensors are being widely used in indoor environments to improve quality of life. In particular, sensors can be installed inside a car to detect dangerous situations, such as a child or a pet being left alone in a car. In this paper, we present an effective method for monitoring people inside a vehicle by using impulse radio ultra-wideband radar. First, we identify that the waveform of the received signal varies with the arrangement of people inside the vehicle. Then, we extract features that represent the statistical characteristics of the received signal and use them as classification criteria. To determine the importance of each feature and reduce the number of features, we use a neighborhood component analysis algorithm. Next, we use ensemble learning with a decision tree as a base classifier to classify the various arrangement of people inside the car. The classification results demonstrate that our proposed method successfully estimates the position and number of people sitting inside a vehicle with an accuracy higher than 90%. Moreover, to utilize the proposed method in real time, we derive an appropriate number of features and structure for the classifier, so as to achieve high classification accuracy while keeping the computational complexity low.