CSIRO, Division of Mathematics, Informatics, and Statistics, Locked Bag 17, North Ryde NSW 1670, Australia.
Microsc Microanal. 2013 Apr;19(2):451-60. doi: 10.1017/S1431927612014328. Epub 2013 Mar 1.
The ability to correctly track objects in time-lapse sequences is important in many applications of microscopy. Individual object motions typically display a level of dynamic regularity reflecting the existence of an underlying physics or biology. Best results are obtained when this local information is exploited. Additionally, if the particle number is known to be approximately constant, a large number of tracking scenarios may be rejected on the basis that they are not compatible with a known maximum particle velocity. This represents information of a global nature, which should ideally be exploited too. Some time ago, we devised an efficient algorithm that exploited both types of information. The tracking task was reduced to a max-flow min-cost problem instance through a novel graph structure that comprised vertices representing objects from three consecutive image frames. The algorithm is explained here for the first time. A user-friendly implementation is provided, and the specific relaxation mechanism responsible for the method's effectiveness is uncovered. The software is particularly competitive for complex dynamics such as dense antiparallel flows, or in situations where object displacements are considerable. As an application, we characterize a remarkable vortex structure formed by bacteria engaged in interstitial motility.
在显微镜应用的许多领域中，正确地在时移序列中跟踪物体的能力非常重要。单个物体的运动通常表现出一定程度的动态规律性，反映出存在潜在的物理或生物学现象。当利用这种局部信息时，可以获得最佳结果。此外，如果已知粒子数量大致保持不变，则可以根据它们与已知最大粒子速度不兼容的情况拒绝大量跟踪场景。这代表了全局性质的信息，理想情况下也应该加以利用。不久前，我们设计了一种有效的算法，该算法同时利用了这两种类型的信息。通过一种新颖的图结构，将跟踪任务简化为最大流最小成本问题实例，该图结构包含代表三个连续图像帧中物体的顶点。这里首次解释了该算法。提供了一个用户友好的实现，并揭示了负责该方法有效性的特定松弛机制。该软件在复杂动态方面特别具有竞争力，例如密集的反平行流，或者在物体位移较大的情况下。作为应用，我们描述了由参与间质运动的细菌形成的一个显著的涡旋结构。
