Radar reflection characteristics of vehicles for contour and feature estimation

Accurate environmental perception is a key requirement for autonomous driving. While the robust and precise estimation of the dynamic state of nearby objects is sufficient for ordinary driver assistance systems like adaptive cruise control, higher levels of autonomy require knowledge of the extent of objects for measurement data association and path finding algorithms. Extent estimation is known to be robustly accomplished by lidar sensors as they provide low measurement noise and a high resolution. Radar-based extent estimation, however, would be more cost-efficient if sufficient robustness would be given. In this work, we will examine the radar characteristics of vehicles using commercial off-the-shelf radars providing cluttered detection data, and evaluate its potential for extent estimation. This is done by observing measurement data from test vehicles with a known position from a spectrum of views, i.e. combinations of distances and orientations. We perform an appropriate measurement analysis and compare the obtainable extent estimate with the true contour. Besides, we consider radar-only vehicle features. The aim of this work is the acquisition of knowledge about typical radar reflection characteristics of vehicles employing a large data basis which can be subsequently utilized in online extent estimation algorithms.