A cepstrum-based, vehicle- and speed-independent road roughness index from inertial sensor data

• A novel roughness profile index is proposed, vehicle and speed-independent by design. • Vehicle dynamics is compensated after being learned from a reference trial. • The index is debiased from vehicle dynamics using cepstral features. • Real-world tests prove higher consistency than existing RPI formulations. Objective road quality estimation is essential for smart infrastructure. Using everyday vehicles with low-cost inertial sensors provides a cost-effective and scalable solution. However, vertical acceleration signals are influenced by vehicle dynamics and speed, producing biased estimates. This issue has been addressed by proposing a novel approach that, using cepstral analysis, separates road-induced and vehicle behaviors. Specifically, thanks to a one-time, model-free calibration, it identifies the latter, providing an unbiased estimate of the first. Last, cepstral features also provide robustness to speed. Validation with naturalistic data across vehicles, drivers, and roads demonstrates accuracy and robustness, highlighting its potential for large-scale road monitoring.