目标跟踪算法综述

目标跟踪一直以来都是计算机视觉领域的关键问题，最近随着人工智能技术的飞速发展，运动目标跟踪问题得到了越来越多的关注.本文对主流目标跟踪算法进行了综述，首先，介绍了目标跟踪中常见的问题，并由时间顺序对目标跟踪算法进行了分类：早期的经典跟踪算法、基于核相关滤波的跟踪算法以及基于深度学习的跟踪算法.接下来，对每一类中经典的跟踪算法的原始版本和各种改进版本做了介绍、分析以及比较.最后，使用OTB-2013数据集对目标跟踪算法进行测试，并对结果进行分析，得出了以下结论：1）相比于光流法、Kalman、Meanshift等传统算法，相关滤波类算法跟踪速度更快，深度学习类方法精度高.2）具有多特征融合以及深度特征的追踪器在跟踪精度方面的效果更好.3）使用强大的分类器是实现良好跟踪的基础.4）尺度的自适应以及模型的更新机制也影响着跟踪的精度.

A Survey of Object Tracking Algorithms

Object tracking has always been a key issue in the computer vision field. Recently, with the rapid development of artificial intelligence, the issue of moving object tracking has attracted more and more attention. This paper reviews the main object tracking algorithms. Firstly, we introduce the common problems in object tracking and classify the object tracking algorithms into three groups:early object tracking algorithms, kernelized correlation filters (KCF) object tracking algorithms, deep learning object tracking algorithms. Then, according to the three groups, we introduce and analyze many famous object tracking algorithms and their following improved versions. Finally, we analyze and compare the performance from many object tracking algorithms using dataset OTB-2013, and conclude that:1) Compared with the optical flow method, Kalman, meanshift and other early algorithms, the tracking speed of KCF-based algorithms are faster and the deep learning based algorithms have higher accuracy. 2) The tracking algorithms with multiple feature fusion or deep features have higher tracking accuracy. 3) Powerful classifiers are the basis for good tracking results. 4) Scale adaptation and updating mechanism also affect tracking accuracy.