Scale change and TTC filter for longitudinal vehicle control based on monocular video

This paper presents a filter algorithm that derives robust input values for longitudinal control of driver assistance systems, based on noisy scale change measurements from a monocular video system. In particular, we calculate filtered scale change and reliable Time-To-Collision (TTC), two values that can be applied for video based Active-Cruise-Control and Forward-Collision-Warning systems. We exploit the fact that TTC is directly related to scale change and the calculation can be performed without accurate distance measurements. To meet the requirements of an automotive environment, where high dynamic driving maneuvers may occur, TTC calculation has to consider relative velocity and relative acceleration between the ego vehicle and the target. We do this by applying an Interacting-Multiple-Model filter, which directly processes the noisy scale change measurements and estimates normalized motion parameters. For better smoothing during accelerated driving situations, we propose an extended IMM configuration. Results presented are based on the evaluation of NCAP test sequences.