An extended target tracker based on structural appearance and improved distribution fields for different scenarios

Abstract An extended target tracker based on structural appearance and improved distribution fields is proposed in terms of current issues that there is no one general tracking algorithm to different scenarios. For extended target with specific geometric structure and pose variation in simple scenarios, in order to get stable tracking point, we construct a structural appearance model to represent target and utilize triangle and line character to obtain tracking point. In complex scenarios with certain texture information and image blurring, to reduce computational complexity and enhance the adaptivity of tracking box, an improved distribution fields is presented in which non-uniform delamination technology is proposed and the BRISK is used to detect key points in tracking box. Numerical experiments on simulation sequences and public database were provided to demonstrate the good performance of this proposed scheme. The conclusion can be drawn that triangle achieves the best stability among geometric graphics and skeleton can reflect the intrinsic geometric structure of the target in simple scenarios. Moreover, non-uniform delamination technology can reduce computational complexity and the BRISK can achieve affine transformation parameters in complex scenarios, so the tracking box can adapt to scale and rotation changes when matching is performed.