The dissertation develops new methods for assessing student participation in long (>1 hour) classroom videos. First, the dissertation introduces the use of multiple image representations based on raw RGB images and AM-FM components to detect specific student groups. Second, a dynamic scene analysis model is developed for tracking under occlusion and variable camera angles. Third, a motion vector projection system identifies instances of students talking.
The proposed methods are validated using digital videos from the Advancing Out-of-school Learning in Mathematics and Engineering (AOLME) project. The proposed methods are shown to provide better group detection, and better talking detection at 59% accuracy compared to 42% for Temporal Segment Network (TSN) and 45% for Convolutional 3D neural network (C3D), and dynamic scene analysis can track participants at 84.1% accuracy compared to 61.9% for static analysis. The methods are used to create activity maps to visualize and quantify student participation.
Group detection, Dynamic participant tracking, AOLME dataset
Level of Degree
Electrical and Computer Engineering
First Committee Member (Chair)
Second Committee Member
Third Committee Member
Fourth Committee Member
SHI, WENJING. "Long-term Human Participation Detection Using A Dynamic Scene Analysis Model." (2023). https://digitalrepository.unm.edu/ece_etds/593