Research On Depth Consistency In Augmented Reality

Augmented reality is a hot topic in the application area of science and information technology.Its technology is the combination between computer-generated virtual information(such as graphic image,audio and etc.)and real objects in real-world scenes.It shows the fused scenes to users,so as to strength their experience and cognition of the real-environment.Augmented reality involves many disciplines including display technology,image identification technology,graphics,machine vision,virtual-actual fusion masking and so on.It's widely used in different areas like entertainment,medical treatment and aerospace.With the demands in the application areas of augmented reality as well as the research and development in some related technologies,users turn their attention on the virtual information produced by computer and on the problems of fusion depth consistency in the real scenes.In today's augment reality system,the processing of the depth fusion just puts virtual information into the two dimension scene images,thereby composing the fusion scene of virtual-actual objects.However,it may lead to the deep relationship of the virtual-actual objects unchanged,lacking the depth effect of the space.Eventually,it's unable to make the users feel the reality of fusion.To the problems of virtual-actual objects fusion above,by analyzing and summarizing current approaches which deal with the relationship of fusion depth in virtual-actual objects,and combining the many disciplines of graphics and computer binocular vision,this present thesis put forwards the depth calculation method which is based on the binocular stereo camera,to solve the depth consistency problem in augmented reality.Firstly,this method is based on chessboard template binocular stereo camera to binocular calibration and obtains the intrinsic and extrinsic parameters.Secondly,in order to real-time obtain the disparity map,the thesis introduces semi-global block matching(SGBM)of dynamic programming and design a method for capturing the disparity map in real-time.Then,the depth information of objects is computed using back projection triangle similarity measure principle in three dimensional scene.Finally,by comparing the z coordinate value of virtual-actual object's pixels,the method judges whether the virtual objects render the scene inside the actual object's contour,so as to settle down problem of the depth consistency.Verified by the experiment,the method for real-time obtaining 3D depth information in the actual-scene is illumination invariant and accuracy.