融合边缘片断与水平集分割的目标跟踪算法

提出了一种融合边缘片断特征与水平集分割的非刚体目标跟踪算法.首先通过Gentle AdaBoost训练器训练得到目标边缘特征字典;然后在每一帧图像中把目标边缘特征字典与图像边缘作Chamfer匹配,通过把具备鉴别能力的图像边缘片断特征嵌入到水平集能量函数中,用来降低背景中边缘信息的干扰.实验结果表明,在非刚体目标局部被遮挡的情况下,仍能准确获得目标轮廓,体现出了该跟踪算法良好的跟踪性能.

Non-rigid object tracking based on the boundary fragment feature and level set segmentation

To track non-rigid objects exactly,conventional methods typically suffer from two problems: 1) fast local shape deformation and 2) occlusion.This paper proposes a novel non-rigid object tracking algorithm based on boundary fragments matching and level set segmentation.Our approach embeds the discriminative object boundary feature learned by Adaboost into the levelset energy function.Both efficiency and adaptability of our algorithm are demonstrated by experimental results.