ANALYZING BODY POSTURES AND SELF-INJURIOUS BEHAVIORS USING A KINECT

Alexis M. Goslen (Dr. Shannon Duvall) Department of Computing Sciences

Visual input devices like web cameras, the Kinect gaming device, and the Leap Motion gaming
device are currently relatively cheap and easy to access. We are studying how best to leverage
these devices for at-home monitoring of children with disabilities. We have three major goals for
this research: to detect Self-Injurious Behaviors (SIBs) like hand-biting or head-banging, to
detect body postures, and to test a mapping of body posture to emotion defined in previous
research. We have developed a program that detects both SIBs and body postures using the
skeleton tracking features with the Kinect, and have tested the accuracy of the interpretation of
these postures. We will present our findings of accuracy and limitations on posture detection
with the Kinect. We will also present our results for the emotion mapping portion of the research.
We are using open-sourced 3D position data collected through other emotion-based experiments
and running the data through our algorithms to see if our system detected body postures
indicative of an emotion. The eventual goal for this project is to create software that uses visual
input of all kinds to detect, predict, and react to emotional states. We can create interactive
spaces in which lights and sounds respond to a child based on his or her emotional needs.
Currently the systems require permanent installation or calibrated cameras. Our system is much
cheaper and easier to use, making it more widely applicable.