计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (24): 152-160.DOI: 10.3778/j.issn.1002-8331.2009-0064

• 模式识别与人工智能 • 上一篇    下一篇

多帧监督的相关滤波无人机目标跟踪

林淑彬,吴贵山,许甲云,杨文元   

  1. 1.闽南师范大学 计算机学院,福建 漳州 363000
    2.闽南师范大学 福建省粒计算及其应用重点实验室,福建 漳州 363000
  • 出版日期:2021-12-15 发布日期:2021-12-13

Multi-frame Surveillance of Correlation Filter in UAV Object Tracking

LIN Shubin, WU Guishan, XU Jiayun, YANG Wenyuan   

  1. 1.School of Computer Science, Minnan Normal University, Zhangzhou, Fujian 363000, China
    2.Fujian Key Laboratory of Granular Computing and Application, Minnan Normal University, Zhangzhou, Fujian 363000, China
  • Online:2021-12-15 Published:2021-12-13

摘要:

目标跟踪是无人机的关键技术之一。无人机目标跟踪容易因相机运动、尺度变化等场景的影响,导致跟踪漂移或丢失。提出一种多帧监督的相关滤波无人机目标跟踪算法,加入多帧信息,根据视图的像差监督响应图变化率,有效地提高跟踪器的识别能力。采用裁剪矩阵引入真实负样本,并加入多个历史帧信息提高滤波器的鲁棒性。采用欧几里德范数定义响应图的像差,通过监督像差的变化防止跟踪漂移,得到目标的准确位置。根据相似度进行目标模型更新。在UAV123和VisDrone2019数据集上与其他算法对比实验。结果显示该算法在相机运动、尺度变化等场景具有良好的跟踪鲁棒性和精度。

关键词: 计算机视觉, 目标跟踪, 无人机, 背景感知相关滤波, 欧几里德范数, 多帧监督

Abstract:

Object tracking is one of the key technologies of UAV. UAV object tracking is easy to cause tracking drift or lose due to the influence of scene such as camera motion and scale change. This paper proposes an algorithm, that is multi-frame surveillance of correlation filter in UAV object tracking. By adding multiple frames of information, monitoring the change rate of the response graph according to the aberration of the view. The tracker’s recognition ability is improved effectively. The clipping matrix is used to import the real negative samples, multiple historical frames information is added to improve the robustness of the filter. The Euclidean norm is introduced to define the aberration of response graph, by supervising the change of the aberration to prevent tracking drift, the exact position of the object is obtained. According to the similarity the object model is updated. Compared with other algorithms on the UAV123 and VisDrone2019 datasets. The results show that the algorithm has favourable tracking robustness and precision in camera motion, scale change and other scenes.

Key words: computer vision, object tracking, Unmanned Aerial Vehicle(UAV), background-aware correlation filters, Euclidean norm, multi-frame surveillance