Millimeter-Wave Radar-Based Vehicle In-Cabin Occupancy Detection Using Explainable Machine Learning

Abstract

This article proposes a method for vehicle in-cabin occupancy detection using millimeter-wave (mmWave) radar and machine learning (ML) scheme, and also proposes a simplified ML model. The proposed method was designed for child presence detection (CPD) because it is easy to implement in vehicles and can detect infants with small reflections without external factors by implementing many features. The ML models we use are extreme gradient boosting (XGBoost) and deep neural network (DNN), which achieve 100% accuracy by using 74 different features, and expected to perform better than conventional models implementing similar detection methods. However, these models are black boxes in structure and have low explainability. Therefore, our method improves the explainability and simplicity of model of XGBoost. We develop a simple approximate model of XGBoost which improves explainability and simplicity of the method. Our method provides a new evaluation algorithm and model saving method to further simplify and speed up the approximate model. The proposed simple model can easily be implemented on low-cost firmware with radar modules.