Vision-based scale-adaptive vehicle detection and tracking for intelligent traffic monitoring

This paper presents a novel real-time scale adaptive visual tracking framework and its use in smart traffic monitoring where the framework robustly detects and tracks vehicles from a stationary camera. Existing visual tracking methods often employ semi-supervised appearance models where a set of samples are continuously extracted around the object to train a discriminant classifier between the vehicle and the background. While proving their advantage, many issues are still to be addressed. One is the tradeoff between high adaptability (prone to drift) and preserving original vehicle appearance (susceptible to tracking loss with appearance changes). Another issue is vehicle scale variations due to perspective camera effects which increase the potential for inaccurate classifier update and subsequently tracking drift. Still, scale adaptability received little attention in discriminant trackers. In this paper we propose a Scale Adaptive Object Tracking (SAOT) algorithm that adapts to scale and appearance changes. The algorithm is divided into three phases: (1) vehicle localization using a diverse ensemble of multiple random projections, (2) scale estimation, and (3) data association where detected and tracked vehicles are correlated. Experimental results demonstrate that our method provides robust tracking in case of significant scale variations and helps alleviate the tracker drift problem.