自适应多策略融合快速KCF跟踪算法  

ADAPTIVE MULTI-STRATEGY FUSION FAST KCF TRACKING ALGORITHM

作　　者：王玲 马胜楠 王鹏[1] Wang Ling;Ma Shengnan;Wang Peng(School of Computer Science and Technology,Changchun University of Science and Technology,Changchun 130022,Jilin,China)

机构地区：[1]长春理工大学计算机科学技术学院,吉林长春130022

出　　处：《计算机应用与软件》2023年第5期207-213,304,共8页Computer Applications and Software

基　　金：中央引导地方科技发展资金吉林省基础研究专项(202002038JC)。

摘　　要：针对KCF算法无法应对跟踪目标尺度变化的问题,提出自适应尺度更新策略,使算法自适应调节窗口尺寸;融合FHOG特征与采用PCA降维后的CN特征,提高KCF算法在复杂背景下的跟踪精度,同时保证改进算法的实时性;通过自适应目标响应策略,使更新的模型更适合目标形变,并利用子网格插值方法代替线性插值方法,减少离散傅里叶变换次数,综合地提高KCF算法的跟踪性能。在OTB2015数据集上进行实验,实验结果表明该方法平均跟踪速度为43.5帧/秒,跟踪精度和成功率比KCF分别提升了11百分点和19.9百分点。该算法在复杂多变的环境中具有良好的识别能力,能够实现快速精准的鲁棒跟踪。Aimed at the problem that the KCF algorithm cannot cope with the change of the tracking target scale,an adaptive scale update strategy is proposed to make the algorithm adaptively adjust the window size.The FHOG feature and the CN feature after PCA dimensionality reduction were combined to improve the tracking accuracy of KCF algorithm in complex backgrounds,and it could ensure the real-time performance of the improved algorithm.Through the adaptive target response strategy,the updated model was more suitable for the target deformation,and the sub-grid interpolation method was used to replace the linear interpolation method to reduce the number of discrete Fourier transforms,which comprehensively improved the tracking performance of the KCF algorithm.Experiments were performed on OTB2015 data set.The results show that the average tracking speed of this algorithm is 43.5 frames per second,and the tracking accuracy and success rate are improved by 11 and 19.9 percentage points respectively compared with those of KCF algorithm.This algorithm has good recognition ability in complex and changeable environment and can achieve fast and accurate robust tracking.

关 键 词：目标跟踪 KCF 尺度更新 特征融合 目标响应 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]