Gait recognition based on angles measurement

Other Title(s)

تمييز المشية اعتمادا على زوايا الحركة

Dissertant

al-Zubi, Ahmad M.

Thesis advisor

al-Ani, Muzhir Shaban

Comitee Members

al-Shalabi, Riyad Farhan
Kanan, Ghassan Ghazi
al-Mashaikhi, Akram Muhammad Uthman

University

Amman Arab University

Faculty

Collage of Computer Sciences and Informatics

Department

Department of Computer Science

University Country

Jordan

Degree

Master

Degree Date

2013

English Abstract

Gait recognition has recently become one of the interesting topics since this field is still immature and poses distinctive characteristics.

Characteristics include easy capturing, require low-cost tools, and do not require any contact with humans.

The reported algorithms in this field require using a stripe to be painted on the person clothes to facilitate the extraction of lines from the person’s lower limb.

They also were reported with low recognition rate.

The work presented in this thesis overcomes the stripe painting difficulty.

The proposed gait recognition framework consists of three stages.

In the first stage, gait signature is extracted from the dynamic change of inclination of the lower body parts; thigh and leg.

First stage consists of background subtraction, silhouette extraction, and gait cycle computation.

In the second stage, the undesirable extra lines in the resulted skeleton image from the first stage are eliminated since these lines do not represent the actual inclination lines of the lower extremity.

Second stage consists of silhouette skeleton, Hough line transform, Filter Lines Algorithm, and feature extraction.

In this work, Filter Lines Algorithm is proposed to filter out the extra lines and select only lines that represent thigh and leg of the walking person.

The third stage uses K-Nearest Neighbor classifier to classify the walking person as unauthorized or authorized.

To test performance of the proposed gait recognition framework, four short videos are recorded for each one of thirty two persons resulting in one hundred and twenty eight videos.

Sixty four videos are used to train the framework while the other sixty four are used to test performance of the proposed framework.

Experimental results indicate an accuracy, recall, and precision of 90.63%, 98.31%, and 90.63%, respectively.

The proposed framework depends on human shape assuming the thigh and leg are within the lower half of human body.

It also assumes that the person walking on a horizontally and smooth surface meaning that both thigh and leg have the same vertical line.

Future work should focus on improving the framework to detect persons walking on uneven and non-smooth surfaces.

Main Subjects

Information Technology and Computer Science

Topics

• Digital electronics

No. of Pages

70

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-522830