吕京国, 白颖奇, 王琛. 一种抗遮掩的卫星视频目标跟踪算法[J]. 武汉大学学报 ( 信息科学版). DOI: 10.13203/j.whugis20210653
引用本文: 吕京国, 白颖奇, 王琛. 一种抗遮掩的卫星视频目标跟踪算法[J]. 武汉大学学报 ( 信息科学版). DOI: 10.13203/j.whugis20210653
Lü Jingguo, Bai Yingqi, Wang chen. Anti-occlusion Target tracking algorithm based on satellite video[J]. Geomatics and Information Science of Wuhan University. DOI: 10.13203/j.whugis20210653
Citation: Lü Jingguo, Bai Yingqi, Wang chen. Anti-occlusion Target tracking algorithm based on satellite video[J]. Geomatics and Information Science of Wuhan University. DOI: 10.13203/j.whugis20210653

一种抗遮掩的卫星视频目标跟踪算法

Anti-occlusion Target tracking algorithm based on satellite video

  • 摘要: 卫星视频中的目标极易受到背景遮挡与淹没,造成外观特征匮乏,最终导致跟踪失败。本文以SiamFC网络为基础,提出一种抗遮掩的卫星视频目标跟踪算法,该算法能够有效解决不同遮掩程度下运动目标的跟踪丢失问题。在目标轻度遮掩时,以模板更新网络得到的前向结果作为当前帧的跟踪结果,并将前向结果设为新的前向模板,获取运动目标的变化特征。在目标重度遮掩时,首先,将前向结果设为新的后向模板,利用后向模板对上一帧影像中的目标进行再次跟踪;然后,采用空间位置预测模块获取目标的历史轨迹特征,得到轨迹结果;最后,以再次跟踪得到的双向验证误差为依据,将前向结果和轨迹结果相结合,得到当前帧的跟踪结果。本文以吉林一号卫星视频数据为例,利用典型运动目标飞机、舰船、车辆进行算法验证。结果表明,该算法可以在目标遮掩时提供较为准确的跟踪结果,跟踪精确率与成功率分别达到76.2%和78.9%,算法跟踪速度为33.9帧/秒(FPS)。该算法可以实现遮掩情况下运动目标的持续跟踪与监测,实际应用中能够有效提高智能交通、防灾救灾、战时监控等方面的应急处理能力。

     

    Abstract: Objectives: Using satellite video data as a target tracking data source can effectively remedy the problems of fixed angle and a small monitoring range of ground surveillance video, thus providing a beneficial supplement for disaster prevention and relief, intelligent transportation, crime tracking and other aspects. However, ground objects in satellite videos are complicated. When the texture, geometry, color and other appearance features of the target are blocked and submerged by other objects, tracking drift or tracking loss will occur in the tracking algorithm. Methods: Therefore, based on the SiamFC network, this paper proposes an anti-occlusion target tracking algorithm based on satellite video, which can effectively solve the tracking loss problem of moving targets under different cover-up degrees. When the target occlusion degree is low, the forward result obtained by template update network is used as the tracking result of the current frame, and the forward result is set as a new forward template to obtain the changing characteristics of the moving target. When the target is highly obscured, firstly, set the forward result to the new backward template, and the backward template is used to repeat the tracking of the target in the previous frame; then, the spatial position prediction module is used to obtain the historical trajectory features of the target and get the trajectory result; finally, the forward result and the trajectory result are combined to get the tracking result of the current frame based on the bidirectional verification error obtained from the repeated tracking. Results: Taking the video data of the Jilin-1 satellite as an example, the algorithm is verified by using typical moving target aircraft, ships and vehicles. The results show that the proposed algorithm can provide more accurate tracking results when the target is occlusion, and the tracking accuracy and success rate reach 76.2% and 78.9%, respectively. The tracking speed of the proposed algorithm is 33.9 FPS. Conclusions: The algorithm can realize the continuous tracking and monitoring of moving targets under occlusion conditions, and can effectively improve the emergency handling ability of intelligent transportation, disaster prevention and relief, wartime monitoring and other aspects in practical application.

     

/

返回文章
返回