Abstract
Accurate camera synchronisation is indispensable for many video processing tasks, such as surveillance and 3D modelling. Video-based synchronisation facilitates the design and setup of networks with moving cameras or devices without an external synchronisation capability, such as low-cost web cameras, or Kinects. In this paper, we present an algorithm which can work with such heterogeneous networks. The algorithm first finds the corresponding frame indices between each camera pair, by the help of image feature correspondences and epipolar geometry. Then, for each pair, a relative frame rate and offset are computed by fitting a 2D line to the index correspondences. These pairwise relations define a graph, in which each spanning cycle comprises an absolute synchronisation hypothesis. The optimal solution is found by an exhaustive search over the spanning cycles. The algorithm is experimentally demonstrated to yield highly accurate estimates in a number of scenarios involving static and moving cameras, and Kinect.