Graduate STEM Fellow Profile

Victor Mejia

Project Title: IMPACT LA: Improving Minority Partnerships and Access through CISE-related Teaching
Thesis: Real-Time Multiple Object Tracking In the Presence of Occlusions
College/University: California State University, Los Angeles
Research Advisor: Dr. Eun-Young, Elaine Kang
Degree Sought: Masters in Computer Science
Department: Computer Science Department
Research Focus: Developing an object tracking algorithm for real-time automated video analysis and integration into an existing H.264 reference
Teaching Partner(s): Gabriel Trejo, Israel Hernandez

Description of Research

In order to develop my object tracking algorithm, I will first develop a robust object detection algorithm. As an aid, I will use information already available from an H.264 decoder in regards to motion, and color in order to detect objects. I will then have to develop a compact description of the general object. This includes (but not limited to) color, current motion vector, and shape features. Once this is done, I will develop a robust tracking algorithm that will have to work (1) in real-time and (2) with object occlusions. This entire process will be integrated into an existing H.264 decoder.

Example of how my research is integrated into my GK-12 experience

Essential to my research is the processing of individual frames as images and the understanding of how computers use the matrix data structure. One activity was designed to help students to understand how a computer represents digital images. This activity accompanied a unit on matrices in Algebra 2. The activity involved a live demonstration where a picture was taken of a volunteer student using a digital camera, and the image was then loaded into a mathematical modeling and visualization tool, MATLAB, and shown to students as a myriad of numbers. Students were able to see that images are represented in multi-dimensional matrices with RGB (Red, Green, and Blue). In addition, in order to introduce the concept of estimating the trajectory of an object in a video sequence, I integrated an activity with a lesson in regression and fitting polynomials to discrete data. Students were first introduced to the concept of the coordinate system on an image, and were given data of an object’s position in a video sequence. Students then had to use the regression feature in their calculators to find the best-fit polynomial and had to predict the object’s next location in the video sequence. Through these two activities I was able to introduce students to my research.