Abstract:

In the existing online object tracking algorithms,tracking deviation commonly occurs when there exists a complex deformation of object appearance. In order to solve this problem,this paper employs robust features to describe the object appearance. First,the perception mechanism of the ventral pathway of human visual cortex is imitated,and C2 feature,which is invariant to position and scale and can distinguish complex shapes,is introduced. Then,a novel online object-tracking model based on a cognitive patch set is put forward to recognize C2 feature. In this model,the importance of a cognitive patch is estimated according to its role in the object recognition,and based on the estimated results,online elimination and update of cognitive patches are realized. Meanwhile,an online object /background classifier is adopted to distinguish new candidate patches,thus solving the problem of the error accumulation resulting from the participation of the background part of the object region in the model adjustment. Simulated results indicate that the proposed method is robust and effective in the presence of complex object deformation and severe occlusion.

Key words: biologically-inspired feature, C2 feature, online object tracking, cognitive patch