A Computer Vision-Based Approach for Biological Phenotyping Using Joint Probability Distribution and ROI Hierarchy Approach
Oral Defence Date:
Prof. Ilmi Yoon, Prof. Rahul Singh
Analyzing behaviors of an animal is an important component in many scientific research es. Frequently, it is a very tedious task that may require repetitive analysis of the simplest behavior that is not interesting for researchers to watch at, but important for their researches. Image Tracker is a tool that was originally motivated to support researchers in San Francisco State University (SFSU) to analyze the behavior of manduca sexta (tobacco hornworm) during its ecdysis stage. While it worked well for what it is designed for, it lacks the flexibility and theoretical foundation to be applied to multiple organisms. The purpose of this project is to take the existing Image Tracker and improve it by eliminating both of its fallacies. To do this, hierarchical system is introduced to make Image Tracker a more flexible system . The new system allows users to design a hierarchy of Regions of Interest (ROI) that defines a species themselves, utilizing joint probability density function for robust tracking. The end result is a flexible tracking framework that can track more objects accurately without any change of code.
analysis, image, tracking, color matching, statistic, analysis framework