Research On The Algorithm Of Obtaining The Depth Map From Static Scene In2D Video

In recent years, people’s requirements of visual effects of traditional mediais improving continuous ly,3D-TV has received a lot of attention. While, if thereare adequate3D video resources is an importance proble m restricting whether the3D-TV can develop rapidly. The traditiona l wa y to get3D video contents isthrough the stereo camera to shoot the scene, whic h put forward higher requestfor the shooting enviro nment and equipme nt. This thesis a ims to use existing2Dvideo, through the method of2D-to-3D, make effective comp leme nt to thepresent3D video resources, to meet the growing needs of people.This thes is mainly stud ies two algorithms for the generation of the depthmap of the statistic scene in2D video. The depth map contents the depthinformation in2D scene; the acquis ition of the depth map is a key eleme nt in theprocess of2D-to-3D. This thesis follows two aspects in detail:Firstly, this thesis describes how to get the depth map through extracting thefocus infor matio n of2D image, and the method to judge whether the ele mentbelo ngs to the foreground ele ment by ca lculating the higher-order statistics ofeach ele ment. Then we adopt a graph-based segmentation technique, computingproportion of the foreground ele ments in each component, estimating w hether itbelo ngs to the foreground region. The depth va lue of foreground regions isassigned to zero constantly, the depth value of background regio ns is assignedaccording to the average va lue of high-order statistics of all ele ments inbackground regions. So, we generate the depth map of the entire image.Second ly, we study in detail how to obtain the depth map by using thegeometr ic infor matio n of2D ima ge. Here, we adopt the method that extracts thevanishing point, that is, the intersection of the direction of the visua ldisappearance. First, we use adaptive canny edge detection to get the edgedetection map, this method can select the high and low thresho ld automatically,make up for the lack of instant thresho ld in the traditio nal a lgor ithm. And then,we carry on Hough trans form to get the ma in straight lines in polar coordinates,and the intersectio n point of these lines is the vanishing point. Fina lly, we makethe ima ge segmentation; assign the depth va lue to each region according to thedistance between the center point of the region and the vanishing point.