Distributed Camera Systems and Image Sensor Clouds

想象一下，在一个足球场里，坐满观众的每个人都在用手机拍摄比赛的视频。不同的观众追随不同的球员，一些人跟着球走，另一些人可能会打开或关闭摄像头，许多人在看台上走动。如果我们能够实时访问这些视频，我们如何检索关于游戏的信息？一方面，我们将从多个角度了解球场所有部分的信息，没有任何事件不会被记录下来。另一方面，我们不知道摄像头在任何时候都在哪里，指向哪些方向，甚至不知道有多少。还会有巨大的冗余(意味着场景中的一个点被许多摄像机捕获)，摄像机之间的视场重叠程度不同，相同的对象在不同的图像中以不同的大小出现的情况，以及许多类似的挑战。

Imagine a soccer stadium in which every person in the capacity crowd is using their cell-phone to capture a video of the game. Different spectators follow different players, some follow the ball instead, others may turn their cameras on or off, and many move about the stands. If we were to have access to these videos in real time, how could we retrieve information about the game? On one hand, we would have information about all parts of the pitch from multiple perspectives and no event would go unrecorded. On the other hand, we do not know where the cameras are at any time, in which directions they point, or even how many there are. There would also be huge redundancy (meaning that a point in the scene is captured by many cameras), variable degrees of overlap in fields-of-view between cameras, instances where identical objects appear at different sizes in different images, and a host of similar challenges.