Human Activity Scripts and Queries for Video Databases

University Of Illinois At Chicago, Chicago IL

Investigators

Abstract

This project will focus on developing methods for analysis and retrieval of human activities in video databases that also can facilitate efficient means for highly compressed storage. For this purpose, it is proposed to develop a novel representation method for human activities. Such a representation is termed as action script and it includes a detailed sequence of human actions that fully describe the 3D motions of all the major body parts. Action scripts can be extended to describe actions of multiple persons as well and could also become an innovative means for highly efficient summarization, retrieval and recognition of complex human activities in video databases. Another objective of this proposal is to develop innovative querying tools that facilitate efficient definition and retrieval of single/multi person activities in video databases. The approach proposed is founded on an extension of an innovative method, recently developed by the PI, for Recognition of human activity by Indexing and Sequencing (RISq). The RISq requires only few video samples and provides fast and robust recognition even in conditions of partial occlusion and varying speeds. It is proposed to further extend the RISq method for the proposed generation and recognition of action scripts of individuals and groups. This is an entirely unexplored domain that could lead to significant scientific contributions in analysis and management of video databases.