基于孪生网络的跟踪算法综述

目标跟踪算法共分为两大类,一类是基于相关滤波的跟踪算法,另一类是基于深度学习的跟踪算法。基于相关滤波的跟踪算法的特点是跟踪速度快,跟踪的精度较低。基于深度学习的跟踪算法的特点是精度较高,但跟踪速度较低。随着研究的深入,深度学习中基于孪生网络的跟踪算法很好地平衡了跟踪速度和精度,既保持了基于深度学习的跟踪算法的优点,又大幅度提高了跟踪速度。首先介绍了基于孪生网络的跟踪算法的工作原理,然后根据基于孪生网络的跟踪算法的发展顺序,分别阐述了不同孪生网路跟踪算法的方法,最后对基于孪生网络的跟踪算法做了总结与展望。

Survey of Tracking Algorithms Based on Siamese Networks

Target tracking algorithms are divided into two categories,one is a tracking algorithm based on correlation filtering,and the other is a tracking algorithm based on deep learning.The correlation filtering-based tracking algorithm is characterized by fast tracking speed and low tracking accuracy.The deep learning-based tracking algorithm is characterized by higher accuracy but lower tracking speed.With the deepening of research,the tracking algorithm based on the twin network in deep learning has well balanced the tracking speed and accuracy,that is,the advantages of the tracking algorithm based on deep learning are maintained,and the tracking speed is greatly improved.This paper introduces the working principle of tracking algorithms based on twin networks,and describes the tracking algorithm methods of different twin networks according to the development order of tracking algorithms based on twin networks,finally concludes the tracking algorithms based on twin networks.