基于改进Haar-like 特征的压缩跟踪算法

针对原始算法特征可能出现的特征无法准确表达目标特性的问题,提出一种改进Haar-like特征的压缩跟踪算法。原始算法利用正负样本训练构造分类器,利用分类器对候选样本判定,得到最高分类器响应样本就是目标。进行重采样以更新分类器为下一帧做准备,对出现的问题,使用了一种新的图像特征来表示目标特性,同时加入一系列策略处理样本,去除那些与目标差异较大的样本,并进行仿真。仿真结果表明:该算法不仅提高了分类器对于正负样本的判别性,也降低了算法的计算复杂度,提高了算法的实时性。

Compressive Tracking Method Based on Improved Haar-like Feature

Real-time robust tracking problem is a great challenge in the tracking area. A compressive tracking method based on improved Haar-like feature is proposed. The original method utilizes the positive and negative samples to train a classifier, then the classifier is used to discriminate the candidate samples. The candidate sample which gets the highest classify score is the target. After that the resample are utilized to update the classifier to get ready for next frame. However the original method has some problems. First, the features selected have too much randomness, so the target cannot be well represented by the features selected in the initial stage. Second, all candidate samples are decided by the classifier which need too much calculation, this will affect the real-time quality. To these problems, this paper uses a new image feature to represent the target and embed some methods to pre-process the samples to remove the samples which have little similarity with the target. This can increase the discriminate power of the classifier and decrease the computational complexity which improves the real-time quality of the method.