Research On Key Technologies Of Motion Analysis Based2D/3D Video Conversion

With the rapid development of computer technology and digital cinema technology, the traditional2D HD videos can’t meet people’s visual needs, which has led to rapid advances in3D technology and made3D videos play an increasingly important role in people’s lives. Existing stereoscopic content production methods are stereo cameras capture and2D/3D conversion.2D/3D conversion is widely used commercially because of its low production cost and short production cycle.This article describes the automatic2D/3D conversion technology and semi-automatic2D/3D conversion technology. For automatic conversion, its key link is depth information extraction; for semi-automatic conversion, its key link is the propagation of depth information. The key links for the two technologies can be addressed by the motion analysis technology. This article describes the following motion analysis methods which are commonly used, including the feature point-based motion analysis and the optical flow method.We introduce the TV-L1optical flow calculation method, and propose a dense optical flow optimization method based on region detection and adaptive support weights, which can detect and repair the optical flow in occlusion areas. Firstly, the forward and backward TV-L1dense optical flow of the image sequence is calculated. The occlusion areas are detected based on the optical flow mismatch measurement. According to the gestalt principles of similarity and proximity, a extrapolate method using adaptive supporting weights in the supporting window is proposed to repair the optical flow maps. By visual and quantitative analysis of the image we can see that the present method can repair the optical flow of occlusion area effectively.This paper proposes a depth information propagation method based on feature points matching technology and image morphing technology. A depth map based adaptive feature point matching algorithm is proposed to address the mismatching problem between feature points. Image morphing technology is used to propagate the depth information of key frames. Experimental results show that the depth maps of non-key frames are efficiently generated automatically.The proposed methods are tested in VS2010and OpenCV2.4.9environment. The methods can refine dense optical flow maps and improve feature point matching accuracy, which has practical significance in video2D/3D conversion.