Files in this item



application/pdfLU-DISSERTATION-2018.pdf (59MB)
(no description provided)PDF


Title:Images and depth for high resolution, low-latency sensing and security applications
Author(s):Lu, Jiajun
Director of Research:Forsyth, David
Doctoral Committee Chair(s):Forsyth, David
Doctoral Committee Member(s):Hoiem, Derek; Lazebnik, Svetlana; Wilson, Andy
Department / Program:Computer Science
Discipline:Computer Science
Degree Granting Institution:University of Illinois at Urbana-Champaign
Abstract:The thesis focuses on using images and depths for high resolution, low latency sensing, and then using these sensing techniques to build security applications. First, we introduce the usefulness of high quality depth sensing, and the difficulty to acquire such depth stream via pure hardware approach. Then, we propose our sensor fusion approach, which combines depth camera and color camera. Chapter 2 puts forward a low cost approach to use a high spatial resolution color stream to help aggressively increase the spatial resolution of the depth stream. Continuing this direction, Chapter 3 proposes to use optical ow to forward warp the depth stream according to a high frequency, low latency CMOS color stream. The warping can create a high frequency, low latency depth stream. In both Chapter 2 and Chapter 3, we show that the improved depth sensing can benefit lots of applications. In Chapter 4, we propose a SafetyNet, which can reliably detecting and rejecting adversarial examples. With the revolutionary SafetyNet architecture and the advanced depth sensing, we can reliably prove to users whether a picture of a scene is real or not. In sum, the thesis focuses on improving sensing technologies and building vision and security applications around the sensing technologies.
Issue Date:2018-12-04
Rights Information:Copyright 2018 Jiajun Lu
Date Available in IDEALS:2019-02-06
Date Deposited:2018-12

This item appears in the following Collection(s)

Item Statistics