基于嵌入式GPU的相关滤波跟踪算法改进及实现

徐卓; 康隽睿; 袁博; 韩冬艳; 党清新

doi:10.15892/j.cnki.djzdxb.2022.04.018

弹箭与制导学报 >

2022 , Vol. 42 >Issue 4: 97 - 103

DOI: https://doi.org/10.15892/j.cnki.djzdxb.2022.04.018

基于嵌入式GPU的相关滤波跟踪算法改进及实现

徐卓 ¹ ,
康隽睿 ¹ ,
袁博 ¹ ,
韩冬艳 ¹ ,
党清新 ²

展开

¹ 西安现代控制技术研究所,西安 710065
² 西安科技大学管理学院,西安 710065

徐卓(1995—),男,陕西渭南人,助理工程师,硕士,研究方向:目标识别。

收稿日期: 2022-03-29

网络出版日期: 2025-05-29

收起

Improvement and Implementation of Correlation Filter Tracking Algorithm Based on Embedded GPU

XU Zhuo ¹ ,
KANG Junrui ¹ ,
YUAN Bo ¹ ,
HAN Dongyan ¹ ,
DANG Qingxin ²

Expand

¹ Xi’an Mordern Control Technology Research Institute, Xi’an 710065, China
² School of Management, Xi’an University of Science and Technology, Xi’an 710065, China

Received date: 2022-03-29

Online published: 2025-05-29

Fold

摘要

为解决目标跟踪任务中所面临的长时遮挡、尺度变化、视场抖动、强实时性需求等关键难点问题,基于相关滤波框架提出了一种多尺度下特征自适应更新的目标跟踪算法,使用多尺度特征对目标进行搜索,自适应更新模板特征实现对遮挡、视场抖动等场景下的目标稳定跟踪。同时在嵌入式GPU上进行算法加速,在保证跟踪精度的同时大幅提升跟踪速度,实现在嵌入式设备上的实时跟踪。最后在OTB数据集及自采视频上进行了测试,结果表明改进后的算法在目标遮挡、尺度变化、视场抖动等场景下均有良好的鲁棒性,同时在嵌入式设备上达到50 帧/s的跟踪速度,满足了目标跟踪任务在嵌入式设备中实现时高复杂度、强实时的需求。

关键词： 目标跟踪; 相关滤波; 嵌入式; 并行计算; 实时跟踪

本文引用格式

徐卓 , 康隽睿 , 袁博 , 韩冬艳 , 党清新 . 基于嵌入式GPU的相关滤波跟踪算法改进及实现[J]. 弹箭与制导学报, 2022 , 42(4) : 97 -103 . DOI: 10.15892/j.cnki.djzdxb.2022.04.018

Abstract

In order to solve the key and difficult problems in the target tracking task, such as long-term occlusion, scale change, field of view jitter, strong real-time requirements, etc., based on the correlation filtering framework, a multi-scale feature adaptive update target tracking algorithm is proposed. Multi-scale features are used to search for targets, and template features are adaptively updated to achieve stable target tracking in scenes such as occlusion and field of view jitter. At the same time, the algorithm is accelerated on the embedded GPU, which considerly improves the tracking speed while ensuring the tracking accuracy, and realizes real-time tracking on embedded devices. Finally, the test is carried out on the OTB dataset and self-collected video. The results show that the improved algorithm has good robustness in the scenes of target occlusion, scale change, field of view jitter, etc. At the same time, it can reach 50 帧/s on embedded devices. The tracking speed meets the requirements of high complexity and strong real-time when the target tracking task is implemented in embedded devices.

Key words： target tracking; correlation filtering; embedded; parallel computing; real-time tracking

[an error occurred while processing this directive]

参考文献

原文顺序 | 文献年度倒序 | 文中引用次数倒序

[1]	卢湖川, 李佩霞, 王栋. 目标跟踪算法综述[J]. 模式识别与人工智能, 2018, 31(1):61-76. DOI

[2]	YAN B, PENG H W, WU K, et al. Lighttrack: finding lightweight neural networks for object tracking via one-shot architecture search[C]// IEEE. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2021: 15180-15189.

[3]	张微, 康宝生. 相关滤波目标跟踪进展综述[J]. 中国图象图形学报, 2017, 22(8):1017-1033.

[4]	HENRIQUES 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.

[5]	WU Y, LIM J W, YANG M H. Online object tracking: a benchmark[C]// IEEE. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2013: 2411-2418.

[6]	DANELLJAN M, KHAN F S, FELSBERG M, et al. Adaptive color attributes for real-time visual tracking[C]// IEEE. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2014: 1090-1097.

[7]	ZHANG K, ZHANG L, YANG M H. Fast compressive tracking[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 36(10): 2002-2015. DOI PMID

[8]	BAO C L, WU Y, LING H B, et al. Real time robust l1 tracker using accelerated proximal gradient approach[C]// IEEE. Proceedings of the 2012 Ieee Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2012: 1830-1837.

[9]	FELSBERG M, BERG A, HAGER G, et al. The thermal infrared visual object tracking VOT-TIR2015 challenge results[C]// IEEE. Proceedings of the IEEE International Conference on Computer Vision Workshops. New York: IEEE, 2015: 76-88.

[10]	卢风顺, 宋君强, 银福康, 等. CPU/GPU协同并行计算研究综述[J]. 计算机科学, 2011, 38(3):5-9.

[11]	肖汉. 基于CPU+GPU的影像匹配高效能异构并行计算研究[D]. 武汉: 武汉大学, 2011.

[12]	张庆科, 杨波, 王琳, 等. 基于GPU的现代并行优化算法[J]. 计算机科学, 2012, 39(4):304-311.

[13]	SANDERS J, KANDROT E. CUDA by example: an introduction to general-purpose GPU programming[M]. Sebastopol: Addison-Wesley Press, 2010.

[14]	DATAL N. Histograms of oriented gradients for human detection[C]// IEEE. Proceedings of the 2005 International Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2005: 886-893.

Options

文章导航