Hardware aware fast depth map generation algorithm

Abstract

This paper proposes a jumped matching depth estimation algorithm to calculate the disparity value in the 3D virtual view synthesis. The matching criterion adopts pixel value similarity, distance similarity and color information similarity as the weighted cost aggregation. In order to reduce the depth discontinuing and then providing the comfort of 3D watching, this paper proposes the disparity prediction method to obtain the disparity value in an object. To reduce the depth discontinuous on scene changed, we proposed a depth stability algorithm to smooth the depth transition on sequential depth maps. The experimental result shows that the method proposed in this paper reduces almost 90% computation time than that of the document [7] and the depth map accuracy of the method is very close to that of other methods in SSIM estimation. Finally, the designed algorithm is implemented in FPGA.