Abstract

Interactive displays are driven by natural interaction with the user, necessitating a computer system that recognizes body gestures and facial expressions. User inputs are not easily or reliably recognized for a satisfying user experience, as the complexities of human communication are difficult to interpret in real-time. Recognizing facial expressions in particular is a problem that requires high-accuracy and efficiency for stable interaction environments. The recent availability of the Kinect, a low cost, low resolution sensor that supplies simultaneous color and depth images, provides a breakthrough opportunity to enhance the interactive capabilities of displays and overall user experience. This new RGBD (RGB + depth) sensor generates an additional channel of depth information that can be used to improve the performance of existing state of the art technology and develop new techniques. The Active Shape Model (ASM) is a well-known deformable model that has been extensively studied for facial feature point placement. Previous shape model techniques have applied 3D reconstruction techniques using multiple cameras or other statistical methods for producing 3D information from 2D color images. These methods showed improved results compared to using only color data, but required an additional deformable model or expensive imaging equipment. In this thesis, an ASM model is trained using the RGBD image produced by the Kinect. The real-time information from the depth sensor is registered to the color image to create a pixel-for-pixel match. To improve the quality of the depth image, a temporal median filter is applied to reduce random noise produced by the sensor. The resulting combined model is designed to produce more robust fitting of facial feature points compared to a purely color based active shape model.

Library of Congress Subject Headings

Human face recognition (Computer science); Interactive computer systems--Design; Computer vision; Deformation potential

Publication Date

11-1-2012

Document Type

Thesis

Department, Program, or Center

Computer Engineering (KGCOE)

Advisor

Cockburn, Juan

Advisor/Committee Member

Kwasinski, Andres

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: TA1650 .B45 2012

Campus

RIT – Main Campus

Share

COinS