Occluded object tracking using object-background prototypes and particle filter

摘要

Object tracking in a real-life scenario is very challenging due to occlusion. State-space models like Kalman and particle filters are well known to handle such a particular problem. The particle filter's performance for solving such a problem depends on two issues - motion model and observation (i.e., likelihood) model. The question remains to exist due to the lack of useful observation and efficient motion models. This article presents an impressive observation model based on confidence (classification) score provided by introducing object-background prototypes based discriminative model. The proposed discriminative model is constructed with the prior knowledge of two classes (i.e., object and background) and tries to discriminate between three categories: an object, background, and occluded part of that object. The existing composite motion model handles the object motion and its scale. We also propose a model update technique that adapts the appearance changes of the object during tracking. We evaluate the proposed method on several challenging benchmark sequences. Analysis of the results concludes that the proposed technique can track fully (or partially) occluded object and the object in various complex environments.