Device-Free Human Presence Detection in Public Transport Using UWB Radar and Naïve Bayes

Occupancy monitoring in public transportation has become a focal point due to its implications for service optimization and passenger comfort. This paper presents a device-free human presence detection method using Ultra-wideband (UWB) radar sensors to estimate seat-level occupancy in buses. The system leverages the impact of human bodies on Radio Frequency (RF) signal propagation, utilizing Angle of Arrival (AoA) and Channel Impulse Response (CIR) data from two ceiling-mounted dual-antenna UWB radars. Seating areas are spatially divided into rows and columns for classification, based on which two classification models are developed: a Naïve Bayes classifier based on AoA data and a Random Forest model based on raw CIR data. Experimental results show that the AoA-based Naïve Bayes model achieves classification accuracies of 67.1% for column prediction and 83.2% for row prediction at a per-pulse level, while consistently outperforming the CIR-based Random Forest models by correctly identifying four out of five evaluated positions when aggregating several predictions over time. The findings suggest that deriving compact and interpretable AoA features from raw CIR data leads to better performance for the seat-level classification task. This approach supports scalable deployment and encourages further exploration of AoA-based methods for complex scenarios such as multi-person localization.