3D frame synchronization detection based on classified epipolar geometry parameters

Nowadays, the stereoscopic 3D (S3D) has become more and more popular. However, sometimes people can still feel uncomfort when watching S3D videos. This phenomenon, to some extent, comes from the non-synchronization of the left and right views. That is, when a S3D videos is shot by two individual cameras, the left frames can´t match exactly with their corresponding right frames in time domain. If the frame synchronization could not be ensured between the left and right views, extra disparity would be introduced due to different motion objects in the video content, which may cause visual fatigue. In this paper, we propose a temporal frame synchronization detection method, which adopts classified epipolar geometry parameters. The epipolar geometry parameters are estimated by classified keypoints pair sets. The keypoints are obtained by SIFT algorithm on two views, and then classified based on the difference about motion features. Experimental results show that, in many different types of scenes, the proposed method works well in the S3D frame synchronization detection.