An experimental comparison of online object-tracking algorithms

Abstract

This paper reviews and evaluates several state-of-the-art online object tracking algorithms. Notwithstanding decades of efforts, object tracking remains a challenging problem due to factors such as illumination, pose, scale, deformation, motion blur, noise, and occlusion. To account for appearance change, most recent tracking algorithms focus on robust object representations and effective state prediction. In this paper, we analyze the components of each tracking method and identify their key roles in dealing with specific challenges, thereby shedding light on how to choose and design algorithms for different situations. We compare state-of-the-art online tracking methods including the IVT,¹ VRT,² FragT,³ BoostT,⁴ SemiT,⁵ BeSemiT,⁶ L1T,⁷ MILT,⁸ VTD⁹ and TLD¹⁰ algorithms on numerous challenging sequences, and evaluate them with different performance metrics. The qualitative and quantitative comparative results demonstrate the strength and weakness of these algorithms.