Infant Detection Using In-Cabin Millimeter- Wave FMCW-MIMO Radar and CFAR Algorithm

Abstract

This paper describes a method for detecting infants in a passenger vehicle by constructing a concise decision tree model from several proposed features. The decision tree model is constructed by approximating the model to overcome the black box structure of machine learning. A type of millimeter-wave radar, Frequency Modulated Continuous Wave - Multiple Input Multiple Output (FMCW-MIMO) radar, is used for detection in a passenger vehicle because it is low cost, non-contact, and insensitive to external factors. The data acquired from the radar is preprocessed to extract features. A previous method proposed to detect the presence/absence of infants at each seat using an approximate decision tree model with about 70 features as input. This method is simple and fast by constructing an approximate decision tree model, but it is still difficult to interpret the large number of features. Therefore, this paper outputs the detection points that can be visualized by preprocessing using the Constant False Alarm Rate (CFAR) algorithm. Approximate decision tree models are constructed using these features, and the accuracy is compared with conventional methods to further improve explainability.