In Partial Fulfillment of the Requirements for the Degree of
Master of Science
Will defend his thesis
Change detection in images of the same scene that are temporally separated is an important task in large number of applications such as video surveillance, remote sensing, medical diagnosis and treatment, and security and video patrolling systems. In temporally separated aerial imagery, the dynamic nature of the outdoor environment presents many challenges in detecting relevant changes. These challenges include changes caused by illumination effect, shadows, non-uniform attenuation or differences in atmospheric absorption. In addition, quality of the captured images can also be significantly different due to sensor noise, camera focus or contrast parameter changes, or camera motion. These challenges can cause change detection system to generate many irrelevant changes. Therefore, the main goal of the change detection models is to reduce as many of these irrelevant changes as possible while preserving the relevant changes. This thesis quantitatively compares the performance of several change detection models for detecting presence of new objects in aerial video sequence. Several types of change detection models under different feature spaces are used. A difference image based model uses different types of distance metrics for detecting changes. An object based changed detection model uses local saliency measures to detect presence of new objects. A fusion model is created by combining the object based model and difference image based model. Finally, an evaluation framework is implemented to compare the results of these change detection models.
Date: Friday, April 20, 2012
Time: 2:00 PM
Faculty, students, and the general public are invited.
Advisor: Prof. Shishir Shah