Vehicle detection and tracking with affine motion segmentation in stereo video

Abstract

This paper proposes a novel method for vehicle detection and tracking based on stereo vision, motion analysis, and road detection. Combining stereo vision with motion analysis makes object detection more appropriate than each individual method. Object tracking with motion estimation, which follows the next object detection starting with an initial solution given by the tracking, improves detection accuracy. Road detection with motion segmentation, considering a parallax in stereo vision as a motion, enables on-road vehicle and obstacle detection. All elements constructing the proposed method are founded commonly on affine motion segmentation. Software and hardware implementations become efficient because their parts share the common principal procedure. Simulation results show the proposed method successfully detects and tracks vehicles on moving background in a complicated scene of downtown.