一个新颖的轮廓线跟踪算法

提出了一个新颖的基于隐马尔科夫模型与光流的轮廓线跟踪算法。曲线描绘由B样条形状空间向量来表达,能够捕捉全局和局部变形。提出应用沿曲线的光流计算来预测曲线在下一帧的位置,在预测曲线的基础上,提出应用隐马尔科夫模型来准确定位曲线的位置。隐马尔科夫模型提供了一种有效的概率手段来融合多种量测特征比如边缘,曲线平滑性,区域灰度或颜色统计信息等等,能够更准确的定位曲线位置。基于仿射形状空间的实验了表明本文所提出算法的有效性。

A Novel Contour Tracking Algorithm

This paper proposes a novel contour tracking algorithm based on Hidden Markov Model (HMM) and optic flow. A vector within B-spline shape space that can accommodate global or local deformation is adopted to represent a curve. Computing optic flow along a curve is described for predicting its position in next frame. HMM is then used to refine the position of curve by taking the predicted curve as an initial contour. HMM provides an efficient probabilistic way incorporating multiple visual cues, such as edges, a curve's smoothness, region grey-level/color statistics, which can locate a curve in image domain accurately. Experiments within affine shape space prove the effectiveness of this novel proposed algorithm.