光电工程, 2010, 37 (5): 12, 网络出版: 2010-09-07
集成轮廓跟踪
Ensemble Contour Tracking
摘要
本文将跟踪看作是二分类问题,提出了一种基于Adaboost 集成学习和快速水平集的轮廓跟踪算法。该方法首先在线地训练一个弱分类器的集合用以区分目标和背景,而通过Adaboost 将集合中的各弱分类器组合成一个强分类器,并用于标定下一帧中的各像素的类别属性,从而确定快速水平集算法的速度函数,然后采用基于动态邻近区域快速水平集来演化目标边界曲线以实现目标的轮廓跟踪。为适应目标和背景的变化,在跟踪过程中在线训练新的弱分类器,而时间相关性则通过更新包含新弱分类器的集合来实现。实验结果表明,在摄像机运动、光照变化、部分遮挡或目标尺度变化等情况下,能实现刚体或非刚体目标的轮廓跟踪。
Abstract
Tracking is considered as a binary classification problem in this paper, and a novel contour tracking algorithm is proposed based on Adaboost ensemble learning and fast level set. First, an ensemble of weak classifiers is trained online to distinguish between the target and the background. Then, the ensemble of weak classifiers is combined into a strong classifier using AdaBoost and the strong classifier is used to label pixels in the next frame as either belonging to the object or the background, so the velocity function of fast level set is obtained. Contour tracking is realized by evolving the zero level set curve using dynamic neighbor region fast level set algorithm which is proposed in this paper. Temporal coherence is maintained by updating the ensemble with new weak classifiers that are trained online during tracking. Experiments show that this algorithm can track the target contour under the conditions of moving background, illumination variation, partial occlusion and the scale change of target.
危自福, 毕笃彦, 徐建军, 南秦博. 集成轮廓跟踪[J]. 光电工程, 2010, 37(5): 12. WEI Zi-fu, BI Du-yan, XU Jian-jun, NAN Qin-bo. Ensemble Contour Tracking[J]. Opto-Electronic Engineering, 2010, 37(5): 12.