Adaptive Extraction of Fused Feature for Panoramic Visual Tracking

Abstract

Panoramic visual tracking is very useful for numerous applications. However, distorted imaging of panoramic vision is prone to affect robustness and lose the target. A panoramic visual tracking method based on adaptive feature fusion is proposed in this paper. Size variation of the target trapezoid box during target movement is labelled. The linear model describing parameter variation of the trapezoid box is fitted. The target trapezoid region is extracted by the model and then refined through the affine transformation. Based on the particle filtering-based tracking framework, the fusion of color and shape is used as the main feature for target tracking. Particle weight is computed using the Bayesian fusion and recursion formula. Experimental results demonstrate the great superiority of the proposed algorithm over other methods in terms of tracking accuracy and anti-occlusion performance, showing that the proposed algorithm can considerably improve target tracking robustness of panoramic vision.