CIR for Rapid and Gradual Deceleration

This study presents a deep learning-based framework for detecting vehicle deceleration patterns using Ultra-Wideband (UWB) Channel Impulse Response (CIR) analysis. Unlike traditional GPS or IMU-based systems, which struggle in GPS-denied environments such as tunnels, the proposed method leverages UWB CIR signal variations to classify two key driving behaviors: rapid deceleration and gradual deceleration. All data were collected from real-world experiments using UWB devices installed on actual vehicles at a professional highway testing site. CIR signals exhibited distinct amplitude and frequency-domain characteristics under different deceleration conditions, and these differences were used as core features in model training. To enhance performance, Doppler shift components were extracted using Fast Fourier Transform (FFT) and integrated with time-domain and time-frequency features. A parallel deep learning model combining GRU and CNN layers achieved 98.10% classification accuracy. This result demonstrates the practical feasibility of using UWB CIR and Doppler-based sensing for reliable vehicle motion recognition in GPS-inaccessible environments, contributing to the development of safer and more intelligent transportation systems.