Mental health disorders are a leading cause of disability in North America and can represent a significant source of financial burden. Early intervention is a key aspect in treating mental disorders as it can dramatically increase the probability of a positive outcome. One key factor to early intervention is the knowledge of risk-markers -- genetic, neural, behavioral and/or social deviations -- that indicate the development of a particular mental disorder. Once these risk-markers are known, it is important to have tools for reliable identification of these risk-markers. For visually observable risk-markers, discovery and screening ideally should occur in a natural environment. However, this often incurs a high cost. Current advances in technology allow for the development of assistive systems that could aid in the detection and screening of visually observable risk-markers in every-day environments, like a preschool classroom. This dissertation covers the development of such a system. The system consists of a series of networked sensors that are able to collect data from a wide baseline. These sensors generate color images and depth maps that can be used to create a 3D point cloud reconstruction of the classroom. The wide baseline nature of the setup helps to minimize the effects of occlusion, since data is captured from multiple distinct perspectives. These point clouds are used to detect occupants in the room and track them throughout their activities. This tracking information is then used to analyze classroom and individual behaviors, enabling the screening for specific risk-markers and also the ability to create a corpus of data that could be used to discover new risk-markers. This system has been installed at the Shirley G. Moore Lab school, a research preschool classroom in the Institute of Child Development at the University of Minnesota. Recordings have been taken and analyzed from actual classes. No instruction or pre-conditioning was given to the instructors or the children in these classes. Portions of this data have also been manually annotated to create groundtruth data that was used to validate the efficacy of the proposed system.
University of Minnesota Ph.D. dissertation. May 2017. Major: Computer Science. Advisor: Nikolaos Papanikolopoulos. 1 computer file (PDF); vii, 121 pages + 2 mp4 video files
A Non-Intrusive Multi-Sensor RGB-D System for Preschool Classroom Behavior Analysis.
Retrieved from the University of Minnesota Digital Conservancy,
Content distributed via the University of Minnesota's Digital Conservancy may be subject to additional license and use restrictions applied by the depositor.