Speed-independent vibration-based terrain classification for passenger vehicles

Abstract

Terrain physical characteristics can have a significant impact on passenger vehicle handling, ride quality, and stability. Here, an algorithm is presented to classify terrain using a single suspension-mounted accelerometer. The algorithm passes a measured acceleration signal through a dynamic vehicle model to estimate the terrain profile, and then extracts spatial frequency components of this estimated profile. A method is introduced to identify and remove terrain impulses from the profile that are caused by ruts and potholes. Finally, a supervised support vector machine is employed to classify profile segments as members of pre-defined classes (such as asphalt, brick, gravel, etc.). The classification algorithm is validated with experimental data collected with a passenger vehicle driving in real-world conditions. The algorithm is shown to classify multiple terrain types with reasonable accuracy at a range of typical automotive speeds. It is also shown that the removal of terrain impulses prior to classification improves classifier performance.