Research on Computing Based on Stochastic Model of Motion Flows in Crowded Scenes

拥挤场景中大量生命体的运动分析为我们提供了一个新的视角。然而，在这样的场景中，通常很难检测和跟踪运动目标，并且分析目标的流动是一项非常具有挑战性的任务。在本研究中，我们着重于随机模型的运动流，物体的形状和强度分布。我们提出了一个系统，用于在拥挤的场景中单独跟踪许多对象。该方法将狄利克雷过程隐马尔可夫模型应用于运动物体的光流场。通过计算自适应环形滤波器输出的二次型来检测每个目标，并利用支持向量机进行分类。原理和实验结果表明，该方法可以对大流量的场景进行分析，并且可以对场景中的每个目标进行单独跟踪。

Motion analysis of huge number of living objects in crowded scenes provides a new insight to us. However, it is generally difficult to detect and track moving objects in such scenes, and analysis of object's flows is a very challenging task. In this research, we focused on stochastic model of motion flows, object's shape and intensity distribution. And we proposed a system for individually tracking many objects in crowded scenes. This method applies Dirichlet Process Hidden Markov Model to the optical flows of moving objects. Each object is detected by calculating quadratic form of adaptive ring filter output and it is classified by support vector machine. The principal and experimental results showed that the huge flows can be analyzed, and each object can be individually tracked in such scenes.