基于孪生神经网络的目标跟踪算法综述

A survey of object tracking algorithm based on Siamese neural network

目标跟踪是计算机视觉中的重要研究课题之一,在智能驾驶、视频监控以及智能医疗诊断等视觉领域得到广泛应用。近年来,基于孪生神经网络的目标跟踪表现出良好的跟踪性能,尤其是在平衡跟踪速度和精确度方面,受到了国内外研究人员的广泛关注。为了更好地结合孪生神经网络进行目标跟踪、充分发挥孪生神经网络在目标跟踪中的优势,本文对基于孪生神经网络的跟踪算法进行了总结与分析。首先着重介绍基于孪生神经网络的目标跟踪算法的基本框架,对现有的基于孪生网络目标跟踪算法进行归类分析,阐述其跟踪原理、创新点以及不足之处。然后在此基础上,采用OTB100数据集进行实验对比,比较多种基于孪生神经网络跟踪算法的性能。实验结果表明,基于孪生神经网络的目标跟踪算法在不同属性下均表现出良好的跟踪性能。最后对基于孪生神经网络跟踪算法的性能进行总结,并对目标跟踪的发展趋势进行展望。

Object tracking is a fundamental research topic in computer vision with wide applications such as intelligent driving,video surveillance and intelligent medical diagnosis.Recently,object tracking based on Siamese neural network presents good tracking performance,especially in balancing tracking speed and accuracy,which has attracted extensive attention of researchers at home and abroad.In order to give full play to the advantages of Siamese neural network in object tracking,the tracking algorithm based on Siamese neural network is summarized and analyzed in the paper.Firstly,we introduce the basic framework of SNN based trackers.Then,we summarize the existing SNN based tracking algorithms,analyzing the tracking principles,advantages and disadvantages.Based on the analysis in details,the performance of tracking algorithms based on SNN is compared on OTB100 benchmark.The experimental results show that the object tracking algorithms based on SNN achieve robust tracking performances under different attributes.Finally,we conclude the performance of SNN based trackers and give the further development trends about SNN based trackers.