Fast Depth Recovery Based On Image Sequences

Under the normal case, image sequences based 3D reconstruction needs stereo matching to obtain dense depth information, which is known as depth recovery. The time consumption of the depth recovery occupies more than 90% of the whole process time of 3D reconstruction, besides, the quality of the recovered depth directly determines the result of the 3D model. Therefore, speeding up the depth recovery rate while maintain the depth quality is significant for the image based 3D reconstruction.One of the main contributions of this paper is our fast depth recovery approach for image sequences. A series of techniques such as sparse sampling, sparse recovery, fast depth diffusion, quick plane fitting and quick bundle optimization are realized to accelerate the original depth recovery method of our laboratory. The accelerated depth recovery algorithm can speed up more than 5 times while maintain the high quality. In addition, to enhance the robustness to changing light sequences, the mutual information is introduced into the depth recovery calculation which lead to a great improvement.Moreover, our paper proposed a fast stereo matching method for binocular sequences with mask information. Making full use of the accurate mask boundary information, we get the initial depth directly by stereo matching of the mask images. Multiple direction scanline optimization are adopted for further refinement. Our approach can get accurate depth information almost real-time.Through a lot of experiments and comparisons, the effectiveness and the high efficiency of our approach have been verified.