Computer Vision

8-23-13

            Dr. Souvenir and his assistance have been working on developing various methods of Computer Vision, the way in which computers detect and analyze their environments.  In this department, Dr. Souvenir and his team have focused specifically on human motion and the ways in which computers can identify the various nuances of how people move through a space and the interactions that occur when two or more people meet.  The goal of this research is to determine patterns and combine this behavior recognition with anthropology in a way to analyze human behavior.  As of now, Computer Vision is capable of person detection and person tracking, but has difficulty with more specific levels of interaction.  These may be low-level interactions such as a handshake or hug, or specific activity recognition such as a group of people in a meeting. 

            The computer vision team is using camera arrays set up in a space to calculate these interactions and movements.   The basic method of determining where a person is in a room is based on creating a background image then searching it for changes.  The greater the number of cameras in an area will yield more accurate results because of the added information.  The cameras triangulate an objects position and look at it through a series of horizontal slices.  This method allows for the cameras and computer to determine the height of an object or person introduced into a space based off of where they exist within the slices.  The difficult part of analyzing people is with all of the subtleties that go into reading body language.  Because body language is such a key part of human communications, the next step in computer vision is to try to create a way to encode human activity and be able to analyze it for recognizable patterns.

            The project has a long way to go, but it’s got a great foundation to build from. Analyzing the way people move through the space gives anthropologists and architects and large amount of information with which to work from.  This data can help architects design spaces in a more responsive manner and aid anthropologists in finding behavioral patterns they may not have noticed with a group of people beforehand.