一种多特征和动态权重策略的目标跟踪算法

传统的单一特征很难准确描述目标的外观特征,无法对目标进行大范围的精确跟踪。针对这一问题,提取卷积特征和颜色直方图特征分别训练相关滤波模型,并提出了一种动态权重策略可自适应地融合两种特征响应图。在模型更新方面,引入目标周围的背景信息,共同训练相关滤波器。将该算法在OTB2013和OTB2015数据集上进行了测试。实验结果表明,该算法可以有效地提高跟踪的准确性和成功率。

Visual Tracking Based on Multi Feature and Dynamic Weight Strategy

The traditional single feature is difficult to accurately describe the appearance characteristics of the target and track the target accurately in a large range.To solve the above problem, the convolution feature and color histogram feature is used for training, and a dynamic weight strategy is proposed to adaptively combine the two functions.In the aspect of model updating, the background information around the target is introduced into the context aware framework as a training sample to jointly train the correlation filter.The algorithm is tested on the OTB2013 and OTB2015 data set.Experimental results show that the algorithm can effectively improve the accuracy and success rate of tracking.