Joshua Young
Research Mentor: Byoung-Keon Park
Mentor Department: University of Michigan Transportation Research Institute, Engineering
Author(s): Not Available
Session: Session 7 (4:00 PM – 4:50 PM)
Presentation Type: Poster 18
Abstract
Accurate and accessible human body models are essential for research in ergonomics, transportation safety, and human–computer interaction, yet existing 3D human meshes often suffer from limited deformation quality and restricted applicability across dynamic scenarios. This project addresses these limitations by improving the realism and usability of open-source human body models and extending their application to behavior analysis. In the first phase, Blender was used to implement and refine linear blend skinning (weight painting) on human mesh models derived from real-time data collected by the University of Michigan Transportation Research Institute and published through HumanShape.org. Particular attention was given to anatomically complex regions—including the hip and groin area, buttocks, and ankle joints—to enable smoother and more realistic joint articulation during interactive movement. In the second phase, the project explored behavioral classification using real-time driving imagery from Hyundai. Leveraging the SAM3D body model, 3D keypoints were extracted from image data, and joint-location features were analyzed to distinguish driving from non-driving behaviors. These features were used to train custom models capable of generalizing across separate image datasets. Together, this work demonstrates how improved mesh deformation techniques and 3D keypoint-based modeling can enhance both the physical realism of human body representations and their utility in applied computer vision tasks, supporting future research in accessible, open-source human modeling and behavior analysis.
