Electronic Theses and Dissertations

Identifier

2609

Date

2016-04-18

Document Type

Thesis (Campus Access Only)

Degree Name

Master of Science

Major

Electrical and Computer Engr

Concentration

Computer Engineering

Committee Chair

Bonny Banerjee

Committee Member

Madhusudhanan Balasubramanian

Committee Member

Aaron L. Robinson

Abstract

Estimation of various anthropometric measurements and body weight is an important problem in medical and forensic domains. In this thesis, we investigate solutions for the challenging problem of weight and height estimation from images and videos, using which we calculate the body mass index (BMI). The proposed approach consists of two steps. First, the measurements of height, chest, waist and hip are obtained from images/videos using a suite of computer vision techniques while the age and gender are assumed to be given. Next, a model is trained to extract the relations between these six variables and the weight, which is then used for inference. A number of models are considered, namely, Gaussian mixture regression, principal component regression, random forest regression, generalized regression neural network, multilayered perceptron, independent component regression, clustering, sparse coding, deep belief network, and copula regression. Performances of the models are compared using root mean square error and correlation coefficient between inferred and actual weights. For data collected by our collaborators as well as other publicly available datasets, a few of the models clearly outperform the rest.

Comments

Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to the local University of Memphis Electronic Theses & dissertation (ETD) Repository.

Download

Share

COinS