Research On Motion Analysis Based Depth Information Estimation Of Vediographics

The progress of modern film and television technology provides audience withmore and better audio-visual experience. The success of3D display technique and3D films inspires people’s enthusiasm for3D audio-visual experience. Meanwhile,3D video industry draws more and more public concern. However, the shooting of a3D movie demands more resources and time, which results in the lack of3D videoresources. Comparatively,2D video resources are rich deposits. If these2D videoscould be transformed into3D videos, the3D video deposits would be enrichedgreatly and the3D movie industry would boom.2D videos to3D videos conversion requires the depth information of thescences. Then, the DIBR(Depth Map Based Rendering) method can be used togenerate the images for left and right eyes.The gist of this thesis is to probe into themethods used to extract the depth information with the motion estimation methodsand the images.Firstly, based on the gray gradient equation of the motion field, iterativeapproximation algorithms are able to give the optimized approximate motion fieldresult under model constraint. The pyramid segmentaton algorithm is used to getbetter approximation. And, motion parameters are taken into computatio n of graygradient. Then, the phase gradient equation of the motion fie ld and a group of Gaborfilters are combined to compute another approximate motion field result. Theweighted average of these two results is set as the final approximation.Secondly, Canny algorithm and region-grow algorithm are carried on toprovide the regional conectivity information, which is used to smooth the motionestimation result. Eyes incline to identify the fast objects as foreground. The morefar away from the camera an object is, the smaller it is displayed. Based on thesetwo principles, objects with higher speed are set as foreground and the depth map iscreated. Finally, the depth map is histogram equalized and smoothed withmorphology method in order to improve the depth map.The depth infromation extraction technique has great applications anddevelopment prospects. On the basis of characteristics of stereo vis ion, modernimage processing techniques are able to obstain the depth infromation from2Dimages. In this thesis, motion information is used to acquire depth information.Experiments present fine results and confirm the practicability of this kind ofmethods.