类孪生网络目标跟踪算法综述

目前,在视觉目标跟踪任务中的主流方法是基于模版匹配的跟踪器,这些方法在目标的分类和边界框的回归上具有很强的鲁棒性,主要可以分为判别相关滤波跟踪器和孪生网络跟踪器,这两种方法都有一个类孪生网络的框架。以孪生网络跟踪器为例,该方法通过模版和搜索区域之间的相关操作确定目标的位置,取得了顶尖的性能表现。近年来,Transformer在计算机视觉领域的发展十分迅速,结合了Transformer的类孪生网络跟踪器在速度和精度方面都远超传统的跟踪方法。文章简要概括了判别相关滤波跟踪器、孪生网络跟踪器的发展,以及Transformer在目标跟踪任务中的应用。

Review of siamese-like object tracking algorithms

The current mainstream methods in visual target tracking tasks are template-matching-based trackers,which are robust in target classification and bounding box regression,and can be mainly divided into discriminative correlation filter trackers and Siamese network trackers,both of which have a Siamese-like framework.Taking the twin network tracker as an example,the method achieves top performance by correlation operations between the template and the search region to determine the location of the target.In recent year,Transformer has been developing rapidly in the field of computer vision,and the Siamese-like tracker incorporating Transformer far exceeds traditional tracking methods in terms of speed and accuracy.This paper briefly summarizes the development of discriminative correlation filter trackers,Siamese network trackers,and the application of Transformer in object tracking.