Moving vehicle detection and tracking using monocular vision

Abstract

This paper aims at real-time vehicle detection and tracking. It presents a novel comprehensive algorithm by multi-cues fusion in sequence images based on monocular vision. Firstly, the vertical symmetry of vehicle rear view is utilised to extract the Region of Interest (ROI) so as to narrow the search area. And then, the sign of underneath shadow is employed to generate hypothetical positions on which potential vehicles maybe present. Following, both image intensity and figure information are used to verify the vertical symmetry of the potential vehicle candidates. Meanwhile, Mean Shift, based on the object features’ model of combine colour Histogram and Orientation Histogram (HOG), is employed to fast search the potential objects. More important, both detection and tracking are under an interactive mechanism which can dramatically improve detection efficiency. Experimental results show this approach can achieve 96.34% accurate rate and run on an average