Please wait a minute...
浙江大学学报(工学版)  2019, Vol. 53 Issue (8): 1488-1495    DOI: 10.3785/j.issn.1008-973X.2019.08.007
计算机与控制工程     
基于多高斯相关滤波的实时跟踪算法
熊昌镇1(),王润玲2,邹建成2
1. 城市道路交通智能控制技术北京市重点实验室,北京 100144
2. 北方工业大学 理学院,北京 100144
Real-time tracking algorithm based on multiple Gaussian-distribution correlation filters
Chang-zhen XIONG1(),Run-ling WANG2,Jian-cheng ZOU2
1. Beijing Key Laboratory of Urban Road Traffic Intelligent Control Technology, Beijing 100144, China
2. School of Sciences, North China University of Technology, Beijing 100144, China
 全文: PDF(2641 KB)   HTML
摘要:

针对分层卷积特征目标跟踪算法实时性不足和单分类器对目标表观变化适应能力差的问题,提出多高斯相关滤波器融合的实时目标跟踪算法. 为了加快跟踪算法,提取VGG-19网络的Pool4和Conv5-3层的多通道卷积特征,通过稀疏采样减少卷积特征通道数;为了防止特征减少造成精确度下降,利用不同高斯分布样本训练多个相关滤波器,并对所有分类器预测的目标位置进行自适应加权融合,提高算法对目标姿态变化的鲁棒性;采用稀疏模型更新策略,进一步提高算法速度,使算法具有实时性. 在OTB100标准数据集上对算法进行测试,结果表明,该算法的平均距离精度为86.6%,比原分层卷积特征目标跟踪算法提高了3.5%,在目标发生遮挡、形变、相似背景干扰等复杂情况时具有较好的鲁棒性;平均跟踪速度为43.7帧/s,实时性更好.

关键词: 视觉跟踪卷积特征相关滤波高斯分布自适应融合    
Abstract:

Aiming at the shortage of real-time performance of the hierarchical convolutional features for visual tracking algorithm and the poor adaptability of single classifier to target appearance changes, a real-time visual tracking algorithm based on multiple Gaussian-distribution correlation filters was proposed. Features with high dimensions of convolution channels were extracted from Pool4 and Conv5-3 layers of VGG-19 networks, and the sparse sampling approach was used to reduce the number of convolution channels to speed up the tracking algorithm. In order to prevent the decrease of tracking accuracy caused by the reduction of features, the multiple correlation filters based on different Gaussian-distribution samples were trained and all the predicted target positions were fused by adaptive weights, expecting for the better robustness for target posture changes. The sparse model update strategy was applied to further improve the algorithm’s speed and achieve the real-time performance. Experimental results on OTB100 benchmark dataset showed that the proposed algorithm had an average distance precision of 86.6%, which was 3.5% higher than that of the original hierarchical convolutional features for visual tracking method. The proposed method has better robustness under complex conditions, for example occlusion, deformation, similar background interferences. The average tracking speed was 43.7 frames per second, and it had a better real-time effect.

Key words: visual tracking    convolutional feature    correlation filter    Gaussian distribution    adaptive fusion
收稿日期: 2018-07-19 出版日期: 2019-08-13
CLC:  TP 391  
作者简介: 熊昌镇(1979—),男,副教授,从事视频分析、深度学习研究. orcid.org/0000-0001-7645-5181. E-mail: xczkiong@163.com
服务  
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章  
熊昌镇
王润玲
邹建成

引用本文:

熊昌镇,王润玲,邹建成. 基于多高斯相关滤波的实时跟踪算法[J]. 浙江大学学报(工学版), 2019, 53(8): 1488-1495.

Chang-zhen XIONG,Run-ling WANG,Jian-cheng ZOU. Real-time tracking algorithm based on multiple Gaussian-distribution correlation filters. Journal of ZheJiang University (Engineering Science), 2019, 53(8): 1488-1495.

链接本文:

http://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2019.08.007        http://www.zjujournals.com/eng/CN/Y2019/V53/I8/1488

图 1  多高斯相关滤波算法框架图
图 2  不同维数卷积特征响应图
图 3  不同高斯样本分布对不同视频图像的特征响应图
数据集 变量 深度特征跟踪器 相关滤波跟踪器
本研究算法 CF2 MSDAT HDT DeepSRDCF KCF SRDCF SAMF DSST Staple
1)注:测试的9种算法速度为对应原算法文献中给出的速度。
OTB 2013 平均DP/% 89.5 89.1 86.3 88.9 84.9 74.0 83.8 78.5 74.0 79.3
平均OP/% 83.8 74.0 74.1 73.7 79.5 62.3 78.1 73.2 67.0 75.4
平均CLE/像素 12.7 15.7 14.6 15.9 25.7 35.5 35.2 30.1 41.2 30.6
平均速度/(帧·s?1 46.2 11.0 23.7 6.3 0.2 273.0 3.6 18.6 26.0 45.0
OTB 100 平均DP/% 86.6 83.7 82.1 84.8 85.1 69.6 78.9 75.1 68.0 78.4
平均OP/% 78.0 65.5 65.5 65.7 77.3 55.1 72.8 67.4 60.1 70.9
平均CLE/像素 15.3 22.8 20.5 20.1 21.4 45.0 38.6 36.5 50.4 31.5
平均速度/(帧·s?1 43.7 10.4 23.5 5.5 0.2 266.0 3.5 17.0 22.0 42.9
表 1  不同算法平均DP、OP、CLE及速度对比1)
图 4  OPE精确度和成功率对比曲线
算法 BC OV IPR FM MB DEF OCC IV SV OPR LR
本研究算法 80.7 60.2 73.8 72.3 75.0 71.4 73.0 79.3 69.1 75.0 60.2
DeepSRDCF 74.9 65.5 71.8 75.5 78.2 69.0 73.7 74.1 73.0 73.8 71.3
CF2 72.1 54.0 66.2 66.8 69.8 60.3 60.6 61.6 51.9 62.9 32.7
HDT 71.3 54.7 65.7 66.4 68.9 61.8 61.1 60.8 51.4 62.7 35.4
MSDAT 72.5 56.0 67.6 63.4 65.9 60.4 59.7 63.5 50.8 63.6 35.9
表 2  不同属性数据集上的成功率对比
图 5  不同策略的精确度和速度对比
图 6  4种算法对具有挑战性的视频序列的跟踪效果对比
1 HERIQUES J F, CASEIRO R, MARTINS P, et al. Exploiting the circulant structure of tracking-by-detection with kernels [C]// Proceedings of European Conference on Computer Vision. Heidelberg: Springer, 2012: 702-715.
2 HERIQUES J F, CASEIRO R, MARTINS P, et al High-speed tracking with kernelized correlation filters[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 37 (3): 583- 596
3 熊昌镇, 赵璐璐, 郭芬红 自适应特征融合的核相关滤波跟踪算法[J]. 计算机辅助设计与图形学学报, 2017, 29 (6): 1068- 1074
XIONG Chang-zhen, ZHAO Lu-lu, GUO Fen-hong Kernelized correlation filters tracking based on adaptive feature fusion[J]. Journal of Computer-Aided Design and Computer Graphics, 2017, 29 (6): 1068- 1074
4 DANELLJAN M, KHAN F S, FELSBERG M, et al. Adaptive color attributes for real-time visual tracking [C]// Proceeding of CVPR 2014. Washington D C: IEEE, 2014: 1090-1097.
5 马晓楠, 刘晓利, 李银伢 自适应尺度的快速相关滤波跟踪算法[J]. 计算机辅助设计与图形学学报, 2017, 29 (3): 450- 458
MA Xiao-nan, LIU Xiao-li, LI Yin-ya Fast scale-adaptive correlation tracking[J]. Journal of Computer-Aided Design and Computer Graphics, 2017, 29 (3): 450- 458
doi: 10.3969/j.issn.1003-9775.2017.03.007
6 BERTINETTO L, VALMADRE J, GOLODETZ S. Staple: complementary learners for real-time tracking [C]// Proceeding of CVPR 2016. Washington D C: IEEE, 2016: 1401-1409.
7 DANELLJAN M, HAGER G, KHAN F, et al. Accurate scale estimation for robust visual tracking [C]// Proceedings of British Machine Vision Conference. Nottingham: BMVA Press, 2014: 1-11.
8 LUKEZIC A, VOJIR T, ZAJC L C, et al. Discriminative correlation filter with channel and spatial reliability [C]// Proceeding of CVPR 2017. Washington D C: IEEE, 2017: 4847-4856.
9 陈倩茹, 刘日升, 樊鑫, 等 多相关滤波自适应融合的鲁棒目标跟踪[J]. 中国图象图形学报, 2018, 23 (2): 269- 276
CHEN Qian-ru, LIU Ri-sheng, FAN Xin, et al Multi-correlation filters method for robust visual tracking[J]. Journal of Image and Graphics, 2018, 23 (2): 269- 276
10 DANELLJAN M, HAGER G, KHAN F S, et al. Learning spatially regularized correlation filters for visual tracking [C]// Proceedings of ICCV 2016. Santiago: IEEE, 2016: 4310-4318.
11 蔡玉柱, 杨德东, 毛宁, 等 基于自适应卷积特征的目标跟踪算法[J]. 光学学报, 2017, (3): 262- 273
CAI Yu-zhu, YANG De-dong, MAO Ning, et al Visual tracking algorithm based on adaptive convolutional features[J]. Acta Optica Sinica, 2017, (3): 262- 273
12 DANELLJAN M, ROBINSON A, KHAN F S. Beyond correlation filters: learning continuous convolution operators for visual tracking [C]// Proceedings of ECCV. Amsterdam: Springer, 2016: 472-488.
13 DANELLJAN M, BHAT G, KHAN F, et al. ECO: efficient convolution operators for tracking [C]// Proceedings of CVPR 2017. Washington D C: IEEE, 2017: 6931-6939.
14 MA C, HUANG J B, YANG X K, et al. Hierarchical convolutional features for visual tracking [C]// Proceedings of CVPR 2015. Washington D C: IEEE, 2015: 3074-3082.
15 WANG X Y, LI H X, LI Y, et al. Robust and real-time deep tracking via multi-scale domain adaptation [C]// Proceeding of IEEE International Conference on Multimedia and Expo. Washington D C: IEEE, 2017: 1338-1343.
16 SIMONYAN K, ZISSERMAN A Very deep convolutional networks for large-scale image recognition[J]. Computer Science, 2014, 34 (2): 1409- 1422
17 WANG L J, OUYANG W L, WANG X G, et al. Visual tracking with fully convolutional networks [C]// Proceedings of ICCV 2015. Santiago: IEEE, 2015: 3119-3127.
18 WU Y, LIM J, YANG M H. Online object tracking: a benchmark [C]// Proceeding of CVPR 2013. Portland: IEEE, 2013: 2411-2418.
19 WU Y, LI M J, YANG M H Object tracking benchmark[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37 (9): 1834- 1848
doi: 10.1109/TPAMI.2014.2388226
20 QI Y K, ZHANG S P, QIN L, et al. Hedged deep tracking [C]// Proceeding of CVPR 2016. Las Vegas: IEEE, 2016: 4303-4311.
21 DANELLJAN M, HAGER G, KHAN F S, et al. Convolutional features for correlation filter based visual tracking [C]// Proceeding of ICCVW. Santiago: IEEE, 2016: 621-629.
[1] 赵燕伟,张健,周仙明,吴耿育. 基于视觉-磁引导的无人机动态跟踪与精准着陆[J]. 浙江大学学报(工学版), 2021, 55(1): 96-108.
[2] 熊昌镇,卢颜,闫佳庆. 基于各向异性高斯分布的视觉跟踪算法[J]. 浙江大学学报(工学版), 2020, 54(2): 301-310.
[3] 王建平 张涛. 移动机器人检测与跟踪系统的研究与设计[J]. J4, 2007, 41(10): 1710-1714.