The depth recovery from multiple optical flow detected by fixational eye movements camera
Yuichiro Deguchi, Norio Tagawa · International Workshop on Advanced Image Technology (IWAIT) 2019 · 2019
Recent advances in computer vision technology are remarkable. Among them, in the field of three-dimensional shape restoration, stereo vision which calculates the depth of an object by the principle of triangulation is the mainstream. Although stereo vison has high depth restoration accuracy, occlusion and change in appearance are likely to occur. Therefore, in this research, we focused attention on the depth restoration method by motion stereopsis with monocular eye which is hard to cause the problem described above. For motion stereo vision, motion parallax between consecutive images needs to be small, and in general, restoration accuracy is low. Therefore, motion stereo vision using many continuous image pairs obtained by camera motion simulating random eye movements called fixational eye movements is performed. By doing this, we aim to improve the depth restoration accuracy by increasing the depth information corresponding to each pixel.