[1]储开斌,朱磊,张继.融合KCF和HOG的改进TLD目标跟踪算法[J].常州大学学报(自然科学版),2022,34(01):60-67.[doi:10.3969/j.issn.2095-0411.2022.01.007]
　CHU Kaibin,ZHU Lei,ZHANG Ji.Fusion of KCF and HOG Based Improvement TLD Target Tracking Algorithm[J].Journal of Changzhou University(Natural Science Edition),2022,34(01):60-67.[doi:10.3969/j.issn.2095-0411.2022.01.007]
点击复制

融合KCF和HOG的改进TLD目标跟踪算法()
分享到：

常州大学学报(自然科学版)[ISSN:2095-0411/CN:32-1822/N]

卷:
第34卷
期数:
2022年01期
页码:
60-67
栏目:
计算机与信息工程
出版日期:
2022-01-28

文章信息/Info

Title:
Fusion of KCF and HOG Based Improvement TLD Target Tracking Algorithm
文章编号:
2095-0411(2022)01-0060-08
作者:
储开斌1; 朱磊2; 张继2
(1.常州大学微电子与控制工程学院,江苏常州213164;2.常州大学阿里云大数据学院,江苏常州213164)
Author(s):
CHU Kaibin1;  ZHU Lei2;  ZHANG Ji2
(1.School of Microelectronics and Control Engineering, Changzhou University, Changzhou 213164, China; 2.Aliyun School of Big Data, Changzhou University, Changzhou 213164, China)
关键词:
目标跟踪;  核相关滤波算法;  跟踪-学习-检测算法;  方向梯度直方图特征
Keywords:
target tracking;  KCF algorithm;  TLD algorithm;  HOG features
分类号:
TP 391.4
DOI:
10.3969/j.issn.2095-0411.2022.01.007
文献标志码:
A
摘要:
针对跟踪-学习-检测(TLD)算法跟踪速度慢,对光照变化鲁棒性差的问题,提出了一种融合核相关滤波(KCF)和方向梯度直方图(HOG)的改进TLD目标跟踪算法。该算法将TLD跟踪模块的中值流跟踪替换为KCF目标跟踪,通过循环矩阵将计算从时域转换到频域,大大降低了计算量,提高了算法的跟踪速度; 再通过提取目标的HOG特征,代替TLD检测模块中随机森林检测器的灰度特征,通过对图像的归一化,降低光照强度对检测的影响,增加了检测器检测成功率,提高了算法的鲁棒性。实验表明,改进TLD算法的一次性通过评估(OPE)精确度达78.7%,成功率达74%; 在光照变化下的OPE精确度和成功率也高出TLD算法15%以上; 测试视频的跟踪速度达TLD算法的2倍,具有较好的实时目标跟踪能力。
Abstract:
Aiming at the problem that the tracking-learning-detection(TLD)algorithm has a slow tracking speed and poor robustness to illumination changes, an improved TLD target tracking algorithm combining kernel correlation filtering(KCF)and directional gradient histogram(HOG)is proposed. This algorithm replaces the median flow tracking of the TLD tracking module with KCF target tracking, and converts the calculation from the time domain to the frequency domain through the cyclic matrix, which greatly reduces the amount of calculation and improves the tracking speed of the algorithm; and then extracts the HOG features of the target instead of the gray feature of the random forest detector in the TLD detection module, the normalization of the image reduces the impact of light intensity on the detection, increases the detection success rate of the detector, and improves the robustness of the algorithm. Experiments show that the one-time pass evaluation(OPE)accuracy of the improved TLD algorithm reaches 78.7%, and the success rate reaches 74%; the accuracy and success rate of OPE under light changes are also 15% higher than that of the TLD algorithm; test video tracking has twice the speed of the TLD algorithm and has good real-time target tracking capabilities.

参考文献/References:

[1]COMANICIU D, MEER P. Mean shift: a robust approach toward feature space analysis[J].IEEE Transactions on Pattern Analysis & Machine Intelligence, 2002, 24(5): 603-619.
[2]BOLME D S, BEVERIDGE J R, DRAPER B A, et al. Visual object tracking using adaptive correlation filters[C]. San Francisco: IEEE Computer Society, 2010.
[3]HENRIQUES J F, RUI C, MARTINS P, et al. High-speed tracking with kernelized correlation filter[J]. IEEE Transaction on Pattern Analysis & Machine Intelligence, 2014, 37(3): 583-596.
[4]HELD D, THRUN S, SAVARESE S. Learning to track at 100 fps with deep regression networks[C]. Santiago: Springer, 2016.
[5]MA C, HUANG J B, YANG X, et al. Hierarchical convolutional features for visual tracking[C]. Santiago: IEEE Computer Society, 2015.
[6]KALAL Z, MIKOLAJCZYK K, MATAS J. Tracking-learning-detection[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2011, 34(7): 1409-1422.
[7]郭巳秋, 张涛, 苗锡奎. 引入样本删除机制的TLD粒子群目标跟踪[J].光学精密工程, 2019, 27(5): 1206-1217.
[8]胡欣, 高佳丽. 基于改进的TLD目标跟踪算法[J].计算机应用研究, 2019, 36(5): 1597-1600.
[9]DALAL N, TRIGGS B. Histograms of oriented gradients for human detection[C]. San Diego: IEEE Computer Society, 2005.
[10]KALAL Z, MIKOLAJCZYK K, MATAS J. Forward-backward error: automatic detection of tracking failures[C]. Istanbul: IEEE Computer Society, 2010.
[11]KALAL Z, MATAS J, MIKOLAJCZYK K. Weighted sampling for large-scale boosting[C]. Leeds: British Machine Vision Association, 2008.
[12]KALAL Z, Matas J, Mikolajczyk K. Pn learning: bootstrapping binary classifiers by structural constraints[C]. San Francisco: IEEE Computer Society, 2010.
[13]OZUYSAL M, CALONDER M, LEPETIT V, et al. Fast keypoint recognition using random ferns[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2010, 32(3): 448-461.
[14]宗家辉. 基于TLD框架的目标跟踪算法的研究[D]. 西安: 西安电子科技大学, 2018.
[15]WU Y, LIM J, YANG M H. Online object tracking: a benchmark[C]. Portland: IEEE Computer Society, 2013.

备注/Memo

备注/Memo:
收稿日期: 2021-10-22。
基金项目: 江苏省高等学校自然科学研究面上项目(19KJB510017); 常州大学科研启动基金资助项目(ZMF18020066)。
作者简介: 储开斌(1975—), 男, 江苏南通人, 硕士, 教授。 通信联系人: 张继(1981—), E-mail: zhangji@cczu.edu.cn
更新日期/Last Update: 1900-01-01