中文摘要：

提出了一种使用稠密SIFT特征进行目标跟踪的算法。该算法首先将表达目标的矩形区域分成相同大小的矩形块,计算每一个小块的SIFT特征,再对各个小块的稠密SIFT特征在中心位置进行采样,建模目标的表达。然后度量两个图像区域的不相似性,先计算两个区域对应小块的Bhattacharyya距离,再对各距离加权求和作为两个区域间的距离。因为目标所在区域靠近边缘的部分可能受到背景像素的影响,而区域的内部则更一致,所以越靠近区域中心权函数的值越大。最后提出了能适应目标尺度变化的跟踪算法。实验表明,本算法具有良好的性能。

英文摘要：

An object tracking method using dense SIFT features is introduced.It is described the object by a rectangle region that is divided into small uniform image patches.For every patch we compute the SIFT feature for characterizing its local characteristics.The object appearance can thus be modeled by dense SIFT features uniformly-sampled on the patch centers.By using the Bhattacharyya distance to measure the dissimilarity between two corresponding patches,we define the distance between two image regions as the weighted sum of the distances of all the corresponding patch pairs.The weighting function is introduced for emphasizing the patches near the region center,because the outer part of the region may be corrupted by background pixels while the inner part is more consistent.Finally,it is developed the tracking algorithm that can adapt to object size change.Experiments show the effectiveness of the proposed method.