基于特征流型的对象跟踪模型

Feature Manifold Tracking

在计算机视觉领域,对象跟踪是一个既重要又极具挑战性的研究课题。跟踪的主要困难体现在物体和背景的外观变化、物体的突发性运动、相机的移动和遮挡。本文提出了一种新的方法来解决对象跟踪问题。采用局部特征集合来表示对象,每一个局部特征组织成特征流形,用来模拟该特征点在各种视角条件下的特征点集合。在跟踪的过程中,我们将更新特征流形集合,来反映对象在运动过程中的变化。同时,通过特征流形和特征点的匹配实现跟踪。相对于传统的特征点与特征点的匹配,特征流形与特征点的匹配方式更加准确。因此,基于特征流形的对象跟踪会取得更加准确的结果。

Object tracking is an important and challenging problem in computer vision.The main difficulties are appearance changing of the object and the background,sudden object motion,camera movement and occlusion.In this article,we propose a novel idea to deal with object tracking.We will organize every feature in the object as a manifold,to simulate all possible viewpoints changing.Then we use all these manifolds to practice our tracking procedure.During our tracking procedures,we update our manifolds,including manifold self-updating,adding new manifolds,and deleting outdate manifolds.Comparing to traditional feature-to-feature matching,our feature-to-manifold matching is more accurate.As a result,the feature manifolds tracking exhibits an outstanding performance.