基于在线多示例提升随机蕨丛的目标跟踪

摘要

为了实现在光线变化、目标形变及背景复杂环境下健壮有效的目标跟踪,提出一种基于在线多示例提升随机蕨丛的目标跟踪方法,通过无限冲激响应(IIR)滤波器实现随机蕨丛分类器的在线增量学习,构建在线随机蕨分类器池,并在在线多示例提升框架下对在线随机蕨进行更新和选取,生成在线多示例提升随机蕨丛分类器,利用该分类器对目标候选区域的采样进行分类以确定目标位置,同时构造正例和负例训练集进行在线增量更新.实验结果表明,复杂环境下,算法具有良好的目标跟踪稳定性.%In order to implement efficient and robust object tracking under the circumstances of variant lighting, changing shape and complicated background, an object tracking algorithm based on online multiple instance boost random ferns was proposed, which used Infinite Impulse Response (IIR) filter to implement online incremental learning for random ferns, the pool of online random fern classifier was constructed, and the random ferns were updated and selected by online multiple instance boosting to generate classifier of online multiple instance boost random ferns. The object was located by classifying samples of object candidate region using the classifier, and positive and negative sets were constructed to online update the classifier. The experiment shows that the proposed method has a good target tracking stability under the complex environment.