A VIEW-DEPENDENT ADAPTIVE MATCHED FILTER FOR LADAR-BASED VEHICLE TRACKING

摘要

LADARs mounted on mobile platforms produce a wealthrnof precise range data on the surrounding objects andrnvehicles. The challenge we address is to infer from thesernraw LADAR data the location and orientation of nearbyrnvehicles. We propose a novel view-dependent adaptivernmatched filter for obtaining fast and precise measurementsrnof target vehicle pose. We derive an analytic expressionrnfor the matching function which we optimize to obtainrntarget pose and size. Our algorithm is fast, robust andrnsimple to implement compared to other methods. Whenrnused as the measurement component of a tracker on anrnautonomous ground vehicle, we are able to track in excessrnof 50 targets at 10 Hz. Once targets are aligned using ourrnmatched filter, we use a support vector-basedrndiscriminator to distinguish vehicles from other objects.rnThis tracker provides a key sensing component for ourrnautonomous ground vehicles which have accumulatedrnhundreds of miles of on-road and off-road autonomousrndriving.