Date of Award

8-12-2015

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Mariofanna Milanova

Abstract

Human Action Recognition in Multimedia, in the context of this dissertation work, is called Human Action Recognition in Videos (HARV). It is a system that used to identify human actions happened in testing videos based on some training video samples. In general, the HARV consists of three stages: human object tracking, feature extraction, and training or action classification. Primarily, the system is composed of two modes: first, the training mode in which the system learns about human actions from training video samples, second, the testing mode that allows this system to recognize unknown human actions in testing videos. In this dissertation, several algorithms are developed to implement the HARV. The primarily goal is to build an accurate and efficient system using different algorithms. In order to achieve this goal, novel algorithms and features are developed using all-frames, sub-frames, and even one-frame of videos for feature extraction to achieve the highest possible accuracy. The developed features involve contour-based extracted from boundary of human objects, silhouette-based extracted from the whole body of human objects, motion-based extracted from all-frames or sub-frames in a video, as well as a combinations of these features. For classification, K-Nearest-Neighbor (KNN) and Support Vector Machine (SVM) are employed with different techniques and setup parameters. The achieved experimental results demonstrated promising performance for human action recognition. Salim Al-Ali May 2015 UALR

Download

Share

COinS