2019/02/11 by Suryansh Kumar, Kumar, Suryansh, Ram Srivatsav Ghorakavi +5 · 2 citations
Computer Science · Engineering · #Advanced Vision and Imaging #Optical measurement and interference techniques #Image Processing Techniques and Applications
paper · pdf · doi:10.48550/arxiv.1902.03791
Recent geometric methods need reliable estimates of 3D motion parameters to\nprocure accurate dense depth map of a complex dynamic scene from monocular\nimages citekumar2017monocular, ranftl2016dense. Generally, to estimate\n\precise measurements of relative 3D motion parameters and to validate\nits accuracy using image data is a challenging task. In this work, we propose\nan alternative approach that circumvents the 3D motion estimation requirement\nto obtain a dense depth map of a dynamic scene. Given per-pixel optical flow\ncorrespondences between two consecutive frames and, the sparse depth prior for\nthe reference frame, we show that, we can effectively recover the dense depth\nmap for the successive frames without solving for 3D motion parameters. Our\nmethod assumes a piece-wise planar model of a dynamic scene, which undergoes\nrigid transformation locally, and as-rigid-as-possible transformation globally\nbetween two successive frames. Under our assumption, we can avoid the explicit\nestimation of 3D rotation and translation to estimate scene depth. In essence,\nour formulation provides an unconventional way to think and recover the dense\ndepth map of a complex dynamic scene which is incremental and motion free in\nnature. Our proposed method does not make object level or any other high-level\nprior assumption about the dynamic scene, as a result, it is applicable to a\nwide range of scenarios. Experimental results on the benchmarks dataset show\nthe competence of our approach for multiple frames.\n