Tracking of Human Body from Multiple Views

动作捕捉系统已广泛应用于游戏、计算机动画、性能评估和监控等领域。然而，大多数现有的动作捕捉系统都需要一些设备附着在人体上。这些设备阻止人体自由行动。为了解决这个问题，我们提出了一种基于图像的方法，使用多个相机视图。这个系统能够捕捉到一个非常快速的动作，比如棒球比赛中的投球。这个成功的结果使我们能够在两个主要领域中应用程序。一个是手势识别。我们将从人体捕获的许多运动参数简化为两个主要参数。将手势识别作为在这两个参数对应的二维平面上进行绘图的字符识别。另一个应用是表演动画。我们通过动作捕捉系统建立了一个动作数据库。在数据库中引用适当的动作可以很容易地制作动画视频。事实上，我们制作了一个四分钟的讲故事的动画。

A motion capture system has been used in many domains such as game, computer animation, performance evaluations and surveillance. Most of the existing motion capture systems, however, need some equipments to be attached to the human body. These equipments prevent the human body from acting freely.To cope with this drawback, we proposed an image-based method using multiple camera views. This proposed system is able to capture a very quick action like a pitching in baseball game. This successful result enables us to make applications in two major domains. One is the gesture recognition. We reduced a number of motion parameters captured from human body into only two principal parameters. The gesture recognition is performed as character recognition of drawing on the two dimensional plane corresponding to the two parameters. Another application is performance animation. We built an action database by our motion capture system. Citation of an appropriate action in the database makes it easy to produce an animation video. In fact, we produced a four minutes story telling animation.