Reconstructing a scene from many images has been an important problem in computer vision. Typically, 2D image feature points are detected, matched, and triangulated to …
Structured light scanners reconstruct 3D information about a scene by projecting a known pattern of light, such as a laser line or time-multiplexed gray code, …
Biological brains are dynamic. Recent advances in electrophysiology and neuroimaging have helped uncover various mechanisms through which brains construct and utilize rich variations in neural …
Automatic recognition and segmentation of objects in images is a central open problem in computer vision. The failure of the segmentation-then-recognition paradigm to produce reliable …
With the rise of computer-aided design and automated fabrication machines, many people bemoan the loss of “hands-on” crafts. This thesis is about assisting "hands-on" users …
With few exceptions, robots today are unable to quickly acquire new manipulation skills in the real world. The modern data-driven approach to skill learning is …
Many types of data observed in high dimensional spaces typically lie on or near low-dimensional manifolds and have much lower intrinsic dimension. Visual data captured …
The field of computer assisted radiology has recently seen an explosion of novel techniques that have been made possible due to advancements in computer algorithms. …
Robotic perception often fails on reflective and transparent surfaces. We will describe a new method of passive RGB sensing which uses a calibrated camera located …
Abstract of Multi-frame edge enrichment with SIFT flow, by Zijun Cui, Degree ScM., Brown University, May 2017. Edge information with high quality are desired by …
The image retrieval (IR) approach to image localization has distinct advantages to the 3D and the deep learning (DNN) approaches: it is seen-agnostic, simpler to …
Progress in deep feedforward neural networks has spawned great successes in many practical applications, but these models struggle to reproduce human-level generalization in tasks that …
This thesis develops new constraints for multiview 3D reconstruction, especially in relative pose estimation and dense reconstruction. Current multiview 3D reconstruction methods rely on two- …
We propose a general-purpose probabilistic framework for scene understanding tasks. We show that several classical scene understanding tasks can be modeled and addressed under a …
Traditional visual odometry approaches often rely on estimating the world in the form a 3D cloud of points from keyframes, which are then projected onto …
We develop new representations and algorithms for three-dimensional (3D) scene understanding from images and videos. To model cluttered indoor scenes, we introduce object descriptors that …
Computer vision problems can be generally divided into three categories: recognition, reconstruction and reorganization. Specifically, reorganization problem is usually considered as a mid-stage process that …
