Recovery of 3D Information from 3D Range Data and from 2D/3D Optical Flow and 3D Scene/Range Flow

点击翻译按钮获取中文摘要

The computation of motion via 2D/3D optical flow and 3D scene/range flow is my long term primary research interest. I have an additional primary interest in in applying these and other computational techniques to measure 3D plant growth from 3D range data in a non-invasive, non-contact manner. Over the next 5 years, my key short term primary research objectives are:***(1) To measure accurate 2D optical flow at occlusion boundaries in real image data. Currently regularization approaches produce flow that tends to "bleed" across occlusion edges (where the images of two differently moving objects meet). We propose to compute the flow inside of closed segmented regions to produce flow that (correctly) abruptly changes at occlusion boundaries.***(2) To measure accurate 3D scene/range flow (3D optical flow on surfaces) on stereo and range image sequences. This type of 3D optical flow measures the 3D motion of visible parts of a scene using depth maps produced from stereo images (scene flow) or directly measured using a range sensor. We hope to produce real time scene/range flow that would be useful in car driving, for example.***(3) To measure 3D quantitative plant growth from 3D range data of growing plants in a non-invasive, non-contact manner. This will allow the quantitative evaluation of a plant's growth over its life cycle. One application would be to quantify the growth rate of wild type and genetically modified plants.