一种具有跟踪外观变化目标能力的均值漂移算法

A Mean Shift Tracker with the Ability to Track Object with Changing Appearance

视角变化往往会引起目标外观特征的变化,基于单一颜色直方图模型的均值漂移跟踪算法往往不能适应这种变化。本文在对连续自适应均值漂移算法深入分析的基础上,提出利用目标外观信息的先验知识,对其建立多个颜色分布模型。每帧跟踪结束后,算法都会根据当前的目标特征和周围环境从多个模型的凸组合中选出最有利于下一帧跟踪的参考模型。实验结果表明,该算法能很好地适应目标外观的变化,且计算代价不大。

Changes in the viewpoint often cause appearance change of object. In this case, mean shift tracker, which is based on single color histogram model, often lost the target. In this paper, we research the CamShift tracker in detail and then proposed to build multiple color-distribution-model for the target accordng to prior knowledge of the object＇s appearance. After tracking in each frame, a model selecting step is carried out to find the best model from the convex group of the reference models for tracking in next frame according to the current object appearances and background appearance in its neighborhood. Experimental results show our tracker can deal with appearance change of object in an elegant manner with low comoutational cost.