arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2311.04749cs.DS
收藏

面向片段上多目标跟踪的在线最小费用环流

Online Min Cost Circulation for Multi-Object Tracking on Fragments

Yanbing Wang, Junyi Ji, William Barbour, Daniel B. Work

AI总结:

针对流式碎片化多目标跟踪数据,本文提出负环抵消算法的在线扩展以求解在线最小费用环流,并给出正确性证明与数值实验验证。

摘要:

多目标跟踪(MOT)或全局数据关联问题通常被处理为图上的最小费用流或最小费用环流问题。尽管已有大量研究致力于提升算法效率,但其中大多数聚焦于批处理问题,即所有数据必须同时可用以构建静态图。然而,随着产生流式数据的应用不断增多,需要一种高效的在线算法来处理输入的流式特性。本文提出了著名的负环抵消算法的在线扩展,用于解决带有流式碎片化数据的多目标跟踪问题。我们为所提出的算法提供了正确性证明,并通过数值实验证明了其效率。

英文摘要:

Multi-object tracking (MOT) or global data association problem is commonly approached as a minimum-cost-flow or minimum-cost-circulation problem on a graph. While there have been numerous studies aimed at enhancing algorithm efficiency, most of them focus on the batch problem, where all the data must be available simultaneously to construct a static graph. However, with the growing number of applications that generate streaming data, an efficient online algorithm is required to handle the streaming nature of the input. In this paper, we present an online extension of the well-known negative cycle canceling algorithm for solving the multi-object tracking problem with streaming fragmented data. We provide a proof of correctness for the proposed algorithm and demonstrate its efficiency through numerical experiments.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑