Deep Learning for Vehicle Deceleration Detection Using UWB CIR and Doppler Shift

Traffic accidents remain a significant challenge in road safety systems, highlighting the need for real-time vehicle motion monitoring technologies. Traditional vehicle-tracking systems, which primarily rely on the global positioning system and inertial measurement unit sensors, face limitations in GPS-denied environments such as tunnels and underground parking. To address these challenges, this study proposes a novel vehicle deceleration classification framework based on Ultra-Wideband (UWB) Channel Impulse Response (CIR) analysis and Doppler Frequency Shift (DFS) analysis. A parallel model architecture is developed to integrate CIR time-domain features, time-frequency representations, and DFS-based velocity estimations. The proposed model was evaluated using real-world driving data collected in tunnel environments. It achieved a classification accuracy of 98.10%, outperforming baseline models using only CIR or IMU data. The experimental results demonstrate that DFS features significantly enhance classification robustness, particularly under strong multipath conditions. These findings validate the feasibility of UWB CIR as a real-time deceleration detection modality for intelligent transportation systems, even in GPS-denied environments.