Adaptive Probabilistic Visual Tracking with Incremental Subspace Update
Adaptive Probabilistic Visual Tracking with Incremental Subspace Update
复制标题
DOI：
10.1007/978-3-540-24671-8_37
复制
发表时间：
2004
期刊：
影响因子：
--
通讯作者：
David A. Ross;Jongwoo Lim;Ming-Hsuan Yang
中科院分区：
文献类型：
--
作者：
David A. Ross;Jongwoo Lim;Ming-Hsuan Yang
视觉跟踪本质上是处理随时间变化的非静态数据流。虽然大多数现有算法能够在受控环境中很好地跟踪对象，但如果对象外观或周围照明发生显著变化，它们通常会失败。这是因为这些视觉跟踪算法是在被跟踪对象的模型对于诸如照明或视点的内部外观变化或外部变化是不变的前提下操作的。因此，大多数跟踪算法在开始建立或学习模型后不会对其进行更新。在本文中，我们提出了一种自适应概率跟踪算法，它通过特征基的增量更新来更新模型。为了在两个视角下跟踪目标，我们使用了一种有效的概率方法来采样带有先验的仿射运动参数，并使用最大后验估计来预测其位置。实验结果表明，该方法在光照、姿态和尺度变化较大的情况下均能较好地跟踪目标，且具有较好的实时性。
Visual tracking, in essence, deals with non-stationary data streams that change over time. While most existing algorithms are able to track objects well in controlled environments, they usually fail if there is a significant change in object appearance or surrounding illumination. The reason being that these visual tracking algorithms operate on the premise that the models of the objects being tracked are invariant to internal appearance change or external variation such as lighting or viewpoint. Consequently most tracking algorithms do not update the models once they are built or learned at the outset. In this paper, we present an adaptive probabilistic tracking algorithm that updates the models using an incremental update of eigenbasis. To track objects in two views, we use an effective probabilistic method for sampling affine motion parameters with priors and predicting its location with a maximuma posterioriestimate. Borne out by experiments, we demonstrate the proposed method is able to track objects well under large lighting, pose and scale variation with close to real-time performance.