融合深度特征的改进KCF行人追踪算法

doi:10.3969/j.issn.1000-1565.2023.06.013

摘要/Abstract

摘要： KCF(kernel correlation filter)算法是目前追踪效果最好的算法之一,但KCF算法无尺度自适应能力且不能有效解决遮挡问题.针对KCF算法的缺陷,设计了1种改进的KCF模型,IKCFMDF(improved KCF mixed with deep feature).该模型先由KCF算法得到目标行人的预测框,然后与通过目标检测得到的所有行人检测框相比较,将交并比最大的检测框作为新的训练目标,从而实现KCF的尺度自适应;引入深度特征代替原KCF使用的HOG(histogram of oriented gradient)特征,以避免HOG特征对行人的姿势改变和颜色信息不敏感的缺陷;在追踪目标丢失时保留其深度特征,与之后视频中行人的深度特征进行对比,如果相似度大于设定阈值,则可判定目标重新出现,从而有效解决KCF的遮挡问题.与原始KCF相比,改进的模型在复杂情况下,对行人追踪的成功率提升了36%,相比主流的神经网络模型也有少量的提升.

关键词: 行人追踪, 改进的KCF模型, 深度特征, 目标检测

Abstract: The KCF(kernel correlation filter)algorithm is currently one of the best tracking algorithms, but it has no scale adaptability and cannot solve the occlusion problem. To solve the problems of the KCF algorithm, an improved KCF model, named IKCFMDF(improved KCF mixed with deep feature), is proposed. This model obtains the prediction frame of the target pedestrian by the KCF algorithm, and then compares it with all the pedestrian detection frames obtained through target detection method, and uses the detection frame with the largest intersection-over-union as the new training target, thus achieving the scale adaptation of KCF. By introducing the deep features instead of the HOG features of KCF,the algorithm can avoid the defects that HOG(histogram of oriented gradient)features are insensitive to pose changes and color information of pedestrians. The deep feature is retained when the tracking target is lost, and then the deep feature is compared with the pedestrians in the subsequent videos. If the similarity is greater than the set threshold, it can determine that the target reappears, thereby the occlusion problem of KCF is solved. Compared with the original KCF, the success rate of pedestrian tracking in complex situations is increased by 36%, and compared with the popular neural network models, the proposed model also achieves improvement.- DOI:10.3969/j.issn.1000-1565.2023.06.013融合深度特征的改进KCF行人追踪算法陈向阳,周扬,杨文柱(河北大学网络空间安全与计算机学院,河北机器视觉工程研究中心,河北保定 071000)摘要:KCF(kernel correlation filter)算法是目前追踪效果最好的算法之一,但KCF算法无尺度自适应能力且不能有效解决遮挡问题.针对KCF算法的缺陷,设计了1种改进的KCF模型,IKCFMDF(improved KCF mixed with deep feature).该模型先由KCF算法得到目标行人的预测框,然后与通过目标检测得到的所有行人检测框相比较,将交并比最大的检测框作为新的训练目标,从而实现KCF的尺度自适应;引入深度特征代替原KCF使用的HOG(histogram of oriented gradient)特征,以避免HOG特征对行人的姿势改变和颜色信息不敏感的缺陷;在追踪目标丢失时保留其深度特征,与之后视频中行人的深度特征进行对比,如果相似度大于设定阈值,则可判定目标重新出现,从而有效解决KCF的遮挡问题.与原始KCF相比,改进的模型在复杂情况下,对行人追踪的成功率提升了36%,相比主流的神经网络模型也有少量的提升.关键词:行人追踪;改进的KCF模型;深度特征;目标检测中图分类号:TP391 文献标志码:A 文章编号:1000-1565(2023)06-0665-08An improved KCF mixed with deep feature for pedestrian trackingCHEN Xiangyang, ZHOU Yang, YANG Wenzhu(Hebei Machine Vision Engineering Research Center, School of Cyber Security and Computer, Hebei University, Baoding 071000, China)Abstract: The KCF(kernel correlation filter)algorithm is currently one of the best tracking algorithms, but it has no scale adaptability and cannot solve the occlusion problem. To solve the problems of the KCF algorithm, an improved KCF model, named IKCFMDF(improved KCF mixed with deep feature), is proposed. This model obtains the prediction frame of the target pedestrian by the KCF algorithm, and then compares it with all the pedestrian detection frames obtained through target detection method, and uses the detection frame with the largest intersection-over-union as the new training target, thus achieving the scale adaptation of KCF. By introducing the deep features instead of the HOG features of KCF,the algorithm can avoid the defects that HOG(histogram of oriented gradient)features are insensitive to pose changes and color information of pedestrians. The deep feature is retained when the tracking target is lost, and then the deep feature is compared with the pedestrians in the subsequent videos. If the similarity is greater than the set threshold, it can determine that the target reappears, thereby the occlusion problem of KCF is solved. Compared with the original KCF, the success rate of pedestrian tracking in complex situations is increased by 36%, and compared with the popular neural network models, the proposed model also achieves improvement.- 收稿日期:2022-03-25 基金项目:河北省自然科学基金资助项目(F2018201060) 第一作者:陈向阳(1977—),女,河南三门峡人,河北大学讲师,主要从事机器视觉方向研究.E-mail:chenxyhbu@163.com 通信作者:杨文柱(1968—),男,河北保定人,河北大学教授,博士,主要从事机器视觉方向研究.E-mail:wenzhuyang@163.com第6期陈向阳等:融合深度特征的改进KCF行人追踪算法

Key words: pedestrian tracking, improved KCF model, deep features, target detection

中图分类号:

TP391

陈向阳,周扬,杨文柱. 融合深度特征的改进KCF行人追踪算法[J]. 河北大学学报(自然科学版), 2023, 43(6): 665-672.

CHEN Xiangyang, ZHOU Yang, YANG Wenzhu. An improved KCF mixed with deep feature for pedestrian tracking[J]. Journal of Hebei University(Natural Science Edition), 2023, 43(6): 665-672.

参考文献

[1] 曹磊,王强,史润佳,等.基于改进RPN的Faster-RCNN网络SAR图像车辆目标检测方法[J].东南大学学报(自然科学版), 2021, 51(1): 87-91. DOI: 10.3969/j.issn.1001-0505.2021.01.012.
[2] 李勇,杨德东,韩亚君,等.融合扰动感知模型的孪生神经网络目标跟踪[J].光学学报, 2020, 40(4): 120-131. DOI: 10.3788/AOS202040.0415002.
[3] LI B, YAN J J, WU W, et al. High performance visual tracking with Siamese region proposal network[C] //2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 18-23, 2018, Salt Lake City, UT, USA. IEEE, 2018: 8971-8980. DOI: 10.1109/CVPR.2018.00935.
[4] ZHU Z, WANG Q, LI B, et al. Distractor-aware Siamese networks for visual object tracking[C] //European Conference on Computer Vision, Cham: Springer, 2018: 103-119. DOI: 10.1007/978-3-030-01240-3_7.
[5] 要小涛,王正勇,石伟,等.基于感知哈希与尺度不变特征变换的快速拼接算法[J].四川大学学报(自然科学版), 2021, 58(3): 89-96. DOI: 10.19907/j.0490-6756.2021.033002.
[6] HENRIQUES J F, CASEIRO R, MARTINS P, et al. High-speed tracking with kernelized correlation filters[J]. IEEE Trans Pattern Anal Mach Intell, 2015, 37(3): 583-596. DOI: 10.1109/TPAMI.2014.2345390.
[7] 王思乐,王铭羽,杨文柱,等.基于时空加权的多特征融合动作识别算法[J].河北大学学报(自然科学版), 2019, 39(1): 93-98. DOI: 10.3969/j.issn.1000-1565.2019.01.016.
[8] 葛雯,史正伟.改进YOLOV3算法在行人识别中的应用[J].计算机工程与应用, 2019, 55(20): 128-133. DOI: 10.3778/j.issn.1002-8331.1901-0318.
[9] LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft COCO: common objects in context[C] //European Conference on Computer Vision, Cham: Springer, 2014: 740-755. DOI: 10.1007/978-3-319-10602-1_48.
[10] LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C] //2017 IEEE Conference on Computer Vision, and Pattern Recognition(CVPR), July 21-26, 2017, Honolulu, HI, USA. IEEE, 2017: 936-944. DOI: 10.1109/CVPR.2017.106.
[11] KINGMA D P, BA J L. Adam: a method for stochastic optimization[C] // 2015 International Conference on Learning Representations, May 7-9, 2015, San Diego, CA, USA, Computer Science, 2015:1-15. DOI: 10.48550/arXiv.1412.6980.
[12] ZHENG L, SHEN L Y, TIAN L, et al. Scalable person re-identification: a benchmark[C] //2015 IEEE International Conference on Computer Vision(ICCV), December 7-13, 2015, Santiago, Chile, IEEE, 2016: 1116-1124. DOI: 10.1109/ICCV.2015.133.
[13] WEINBERGER K Q, SAUL L K. Distance metric learning for large margin nearest neighbor classification[J]. J Mach Learn Res, 2009, 10: 207-244. DOI: 10.1007/s10845-008-0108-2.
[14] WU Y, LIM J, YANG M H. Object tracking benchmark[J]. IEEE Trans Pattern Anal Mach Intell, 2015, 37(9): 1834-1848. DOI: 10.1109/TPAMI.2014.2388226.
[15] JIA X, LU H C, YANG M H. Visual tracking via adaptive structural local sparse appearance model[C] //2012 IEEE Conference on Computer Vision and Pattern Recognition, June 16-21, 2012, Providence, RI, USA, IEEE, 2012: 1822-1829. DOI: 10.1109/CVPR.2012.6247880.
[16] KALAL Z, MIKOLAJCZYK K, MATAS J. Tracking-learning-detection[J]. IEEE Trans Pattern Anal Mach Intell, 2012, 34(7): 1409-1422. DOI: 10.1109/TPAMI.2011.239.
[17] HOWARD A, SANDLER M, CHEN B, et al. Searching for MobileNetV3[C] //2019 IEEE/CVF International Conference on Computer Vision(ICCV). October 27 - November 2, 2019, Seoul, Korea(South), IEEE, 2020: 1314-1324. DOI: 10.1109/ICCV.2019.00140.
[18] HUANG R, PEDOEEM J, CHEN C X. YOLO-LITE: a real-time object detection algorithm optimized for non-GPU computers[C] //2018 IEEE International Conference on Big Data(Big Data), December 10-13, 2018, Seattle, WA, USA, IEEE, 2019: 2503-2510. DOI: 10.1109/BigData.2018.8621865.
[19] ZHANG X Y, ZHOU X Y, LIN M X, et al. ShuffleNet: an extremely efficient convolutional neural network for mobile devices[C] //2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 18-23, 2018, Salt Lake City, UT, USA, IEEE, 2018: 6848-6856. DOI: 10.1109/CVPR.2018.00716.
[20] WU B C, DAI X L, ZHANG P Z, et al. FBNet: hardware-aware efficient ConvNet design via differentiable neural architecture search[C] //2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR), June 15-20, 2019, Long Beach, CA, USA, IEEE, 2020: 10726-10734. DOI: 10.1109/CVPR.2019.01099. (