Synchronization of vision based sensor networks with variable frame rates

This paper introduces a synchronization methodology for distributed vision based sensor networks with clock skew or variable frame rates. The methodology requires a ballistic or otherwise predictive object to be tracked by the sensor network and used to calibrate the clock skew and/or relative variable frame rates between the cameras. The relative time stamp of each image captured can be extracted using the dynamic model of the predictive object. The time stamps and a best fit correlation is used to synchronize all the cameras (and their video streams) that are tracking the same predictive object. In sport the predictive object is likely to be a ballistic object such as a ball or a player in flight, while in security applications, the predictive object may be trains, cars or other objects traveling at predictable speeds in known locations.