基于特征优化的稳定目标跟踪方法

A Stable Target Tracking Method Based on Feature Optimization

目标跟踪是视频分析领域中的重要分支,其中核相关滤波(Kernel Correlation Filter,KCF)算法在跟踪精确率、成功率和跟踪性能方面具有不错的表现。但是,在某些场景里,光照变化和形态变化较多会存在目标丢失、耗时长的问题。针对此问题,设计了基于特征优化的稳定目标跟踪方法。考虑目标运动过程的尺寸变化,利用局部二值模式(Local Binary Patterns,LBP)特征替换FHOG特征和初始框动态更新策略,解决了跟踪目标尺寸骤变导致的目标丢失和跟踪耗时长等问题。经过对比数据集OTB100和VOT2018,结果表明本文提出的算法在精确率和稳定性等方面都优于经典KCF算法,且在各种复杂场景下更加准确和稳定,跟踪速度达到180 fps。

Object tracking is an important branch of video analysis.Kernel Correlation Filter(KCF)is a classical correlation filter tracking algorithm.It is a single target tracking algorithm with high accuracy and less time-consuming.However,in some scenes,when there are more changes in illumination and morphology,there will be problems of target loss and less time-consuming.In this paper,aiming at the shorcomings of KCF target tracking algorithm,a stable target tracking method based on feature optimization is designed.Considering the size change of the target movement process,by replacing FHOG feature with Local Binary Patterns(LBP)feature and the dynamic update strategy of initial frame,the problems of tracking loss caused by large change of tracking target size and slow tracking speed of moving target are solved.After comparing the data sets OTB100 and VOT2018,the results show that the tracking performance of the proposed algorithm is better than that of the original KCF algorithm.It is more accurate and stable in various complex scenes,and the tracking speed reaches 180 fps.