Multiview Stereoscopic Imaging System With Binocular Cameras

Free viewpoint television as the most important multi-view imaging applicationsprovides users the ability to interactively select scene viewing angle, breaking therestrictions which can only be viewed from the camera viewpoint, and enhance theuser experience provided by traditional media. Now it is more and more widespreadconcerned.This paper first introduces the development of free viewpoint televisiontechnology at home and abroad, and then proposed a minimum distortion rectificationand background modeling as the pre-processing method, the disparity map isgenerated by local stereo matching algorithm based on the adaptive cross window,and finally generate intermediate virtual viewpoints by linear interpolation, and thenset up a multiview stereoscopic imaging system with binocular cameras, made majorachievements are as follows:The system proposes a new rectification method for uncalibrated stereo images toreduce geometric distortion. Both homographies are restrained to close to rigid onesin the image level and introduce a penalty term to further reduce the distortion in thepixel level. The image rectification problem is formulated into the minimization of anenergy function without explicit computation of fundamental matrix. Use RPCA toextract the background and then the system simply calculates the disparity of theforeground and improved the stability of the disparity between frames. Then thesystem applies a fast stereo matching algorithm based on the cross support window.Then the algorithm aggregates the costs in an adaptive support region based on thecross to achieve the disparity accurately. Its accuracy is compared to global methodsand it retains the advantage of the local algorithm, and finally the system uses thelinear interpolation to generate virtual viewpoint.Experimental results show that the system can generate intermediate virtualviewpoint in real-time for the captured videos and support switching freely betweeneight viewpoints with naked eyes, and avoid the demand of hardware requirements toobtain a three-dimensional video source which greatly reducing the cost.