Vehicle Tracking on Video Sequences via Subspace Learning

Abstract

In this study, we introduce a tracking algorithm which tracks the vehicles marked by an operator on video sequences. Fast Principal Component Pursuit algorithm is used to obtain the background and foreground models with high precision. For foreground model obtained by subspace learning, a thresholding method is proposed. For the detected foreground objects, features are extracted to separate the vehicle marked by the operator from other foreground objects. Tracking of the marked object is performed by Kalman filter. The experimental results show that the proposed method is effective against dynamic backgrounds, complex scenes and occlusions.