Statistical Relational Learning (SRL) as an Enabling Technology for Data Acquisition and Data Fusion in Video.

摘要

While our ability to gather vast amounts of video data is growing at a staggering rate, our ability to effectively store, process, and analyze this video has not kept pace. It is therefore necessary to develop automatic methods for allocating limited resources in video understanding. It particular, it is important to reason about which portions of video require expensive analysis and storage. This project aims to make these inferences using new and existing tools from Statistical Relational Learning (SRL). SRL is a recently emerging technology that enables the effective integration of statistical or probabilistic information, with relational or logical domain information, providing the ability to reason collectively about large, complex, interacting domains.