Spatio-temporal depth data reconstruction from a subset of samples

High-quality depth data is needed in many advanced computer vision as well as 3D and virtual reality applications. To surpass the hardware limitations, computational approaches are commonly exploited, and the solutions are from the intersection of two fundamental problems, depth map super-resolution and inpainting, leading to a general problem of reconstructing depth data from a subset of samples. Extending our previous work [1], we propose a spatio-temporal depth reconstruction (STDR) algorithm, which is scalable to temporal volume. We also present an updated parameter tuning approach and a speed-up scheme for depth video reconstruction application. Experimental results show that the proposed STDR algorithm outperforms the existing methods and is robust to varying temporal volumes.