On-road vehicle detection during dusk and at night

Abstract

The video-based on-road detection of vehicles at daytime allows driver assistance systems to avoid collisions and thereby improve safety, and realize comfort functions, like the well known adaptive cruise control. However, at nighttime, common video sensor based vehicle detection algorithms can't be used, because most state-of-the-art features, like shadows, symmetry and others, cannot be measured. The on-road detection of vehicles at night is an obligatory feature for modern driver assistance systems, because those systems have to provide assistance functionality at day-time and at night-time, either. In this work, vehicles in front of the own car are recognized by detection of their front or rear lights, using a perspective blob filter and subsequently searching for corresponding light pairs. For preceding vehicles, the activity of the third break light is estimated, to distinguish the maneuver state of the vehicle. Experiments show the robustness of the approach during dusk and at night sequences.