College of Physics and Electronic Engineering,Shanxi University,Taiyuan 030006,China
Show less
文章历史+
Received
Published
2023-01-05
2024-02-29
Issue Date
2025-11-27
PDF (4710K)
摘要
空间正则化相关滤波算法跟踪过程中仅采用手工特征表征目标,高斯-赛德尔方法训练滤波器的复杂度高,跟踪结果不可靠时仍逐帧更新模型,导致跟踪效果不佳。针对空间正则化相关滤波算法存在的问题,提出深度特征目标感知交替方向乘子法(Alternating Direction Method of Multipliers,ADMM)优化多指标更新相关滤波跟踪算法。该算法融入预训练网络提取的深度特征,并依回归损失的梯度信息进行通道选择,增强了对目标的表征能力;采用交替方向乘子法训练相关滤波器,降低算法复杂度,提升跟踪速度;根据多指标更新方法判断是否进行模型更新,不但提升了算法运行效率,而且避免了因学习到错误信息而导致的模型腐败。实验结果表明,所提算法的成功率、精确度在数据集OTB2015上均优于其它8种对比算法,且在复杂场景下具有更强的跟踪鲁棒性。
Abstract
A novel algorithm named as correlation filtering algorithm based on target-aware deep tracking by alternating direction method of multipliers(ADMM) optimization and multi-index update is proposed to avoid the disadvantages of only using hand-crafted features to characterize the target, the higher complexity of Gauss-Seidel method, which is used to train the filter, and updating the model frame by frame even when the tracking results are unreliable in SRDCF algorithm. After fusing the deep features that are extracted by the pre-training network, the channel selection is performed according to the gradient information of regression loss, which enhances the characterization ability of the target. In order to reduce the computational complexity and accelerate the tracking speed, the proposed algorithm utilizes the alternating direction method of multipliers to train the correlation filter. Furthermore, whether the model is updated or not depends on the multi-index update method, which not only improves the tracking efficiency, but also avoids the model corruption that is caused by learning wrong information. The experimental results show that the success rate and the precision of the presented algorithm outperform the compared algorithms on the OTB2015 dataset. Under the complex scenarios, the proposed approach has stronger tracking robustness compared with the state of the art approaches.
针对SRDCF算法存在的问题,提出深度特征目标感知ADMM优化多指标更新相关滤波跟踪算法(Target-Aware Deep Tracking by ADMM Optimization and Multi-Index Update, TACF)。该算法融入预训练的深度特征,并依回归损失的梯度信息进行通道选择,提高对目标的表征能力;采用交替方向乘子法(ADMM)[14]训练相关滤波器,降低算法复杂度,加快跟踪速度;根据多指标更新方法判断是否进行模型更新,不但提升了算法运行效率,还避免了因学习到大量干扰信息而导致的模型损坏。实验结果表明,TACF算法相较于其它8种现有算法实现了更精确、更鲁棒的跟踪。
ZHANGChanggong, YANGHaitao, WANGJinyu,et al.Survey on visual single object tracking based on deep learning[J].Application Research of Computers,2021,38(10):2888-2895. (in Chinese)
HENRIQUESJ F, CASEIROR, MARTINSP,et al.Exploiting the circulant structure of tracking-by-detection with kernels[C]//Proceedings of the 12th European Conference on Computer Vision-Volume Part IV,2012:702-715.
[6]
HENRIQUESJ F, CASEIROR, MARTINSP,et al.High-speed tracking with kernelized correlation filters[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(3):583-596.
[7]
LIANGY, WUQ, LIUY,et al.Deep correlation filter tracking with shepherded instance-aware proposals[J].IEEE Transactions on Intelligent Transportation Systems,2022,23(8):11408-11421.
[8]
SUNS, GONGJ, LIY,et al.A research of target tracking algorithm based on deep learning and kernel correlation filter[C]//International Joint Conference on Neural Networks (IJCNN),2022:1-8.
[9]
HOUY, LINX, LIJ.Correlation filter and deep siamese network hybrid algorithm for visual object tracking[C]//IEEE 6th International Conference on Intelligent Computing and Signal Processing (ICSP),2021:73-76.
[10]
CHAOM, HUANGJ B, YANGX,et al.Hierarchical convolutional features for visual tracking[C]//IEEE International Conference on Computer Vision,2015:3074-3082.
[11]
QIY, ZHANGS, QINL,et al.Hedged deep tracking[C]//IEEE Conference on Computer Vision and Pattern Recognition,2016:4303-4311.
[12]
VALMADREJ, BERTINETTOL, HENRIQUESJ F,et al.End-to-end representation learning for correlation filter based tracking[C]//IEEE Conference on Computer Vision and Pattern Recognition,2017:5000-5008.
WANGN, SONGY, MAC,et al.Unsupervised deep tracking[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition,2019:1308-1317.
[15]
DANELLJANM, HAGERG, KHANF S,et al.Learning spatially regularized correlation filters for visual tracking[C]//IEEE International Conference on Computer Vision,2015:4310-4318.
[16]
BOYDS, PARIKHN, CHUE,et al.Distributed optimization and statistical learning via the alternating direction method of multipliers[J].Foundations and Trends in Machine Learning,2010,3(1):1-122.
[17]
SELVARAJUR R, COGSWELLM,DAS A,et al.Grad-CAM:visual explanations from deep networks via gradient-based localization[C]//IEEE International Conference on Computer Vision,2017:618-626.
[18]
WANGM, LIUY, HUANGZ.Large margin object tracking with circulant feature maps[C]//IEEE Conference on Computer Vision and Pattern Recognition,2017:4800-4808.
[19]
SIMONYANK, ZISSERMANA.Very deep convolutional networks for large-scale image recognition[J].arXiv Preprint arXiv:2014.
[20]
WUY,LIM J, YANGM H.Object tracking benchmark[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(9):1834-1848.
[21]
LIY, FUC, DINGF,et al.AutoTrack:towards high-performance visual tracking for UAV with automatic spatio-temporal regularization[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition,2020:11920-11929.
[22]
LIX, MAC, WUB,et al.Target-aware deep tracking[C]//EEE/CVF Conference on Computer Vision and Pattern Recognition,2019:1369-1378.
[23]
LIF, TIANC, ZUOW,et al.Learning spatial-temporal regularized correlation filters for visual tracking[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition,2018:4904-4913.
[24]
XUT, FENGZ H, WUX J,et al.Learning adaptive discriminative correlation filters via temporal consistency preserving spatial feature selection for robust visual tracking[J].IEEE Transactions on Image Processing,2019,28(11):5596-5609.