Accurate Camera Syncronization Using Deep-Shallow Mixed Models

To synchronize multiple cameras has been attracting more and more attentions since the surge of 360-degree view videos and VR/AR applications. Usually only the professional cameras with dedicated external-trigger interface can be synchronized easily and accurately. However, those cameras suffer from the high price and the complex operating procedure. For network cameras, some researchers have proposed synchronization method by using the NTP protocol. To our best knowledge, there is no effective method for synchronizing consuming cameras. In this work, we propose a method for synchronizing nearly all kinds of cameras given that they are connected into on PC. The experiment show that for average consuming webcams, our method can achieve the mean absolute difference (MSD) below 12 ms, which is totally meet the requirement of stereo vision or other multiple vision tasks.