Poster Title (Current Submission)

New Capabilities for Vison-based Posture Prediction

Major(s)

Engineering

Minor(s)

Spanish

Mentor Name

Timothy Marler, Karim Abdel-Malek

Other Mentor Department

Biomedical Engineering

Presentation Date

March 2010

Abstract

Although field of view (FOV) is a commonly used evaluation parameter with digital human models, minimal research has involved modeling how eye motion (relative to the head and body) affects the FOV and posture of a digital human striving to see a particular target. Few models incorporate independent eye movement and the effects of obstacles, with the ability to predict human posture realistically. This work presents two new and critical components for simulating how vision affects human posture: 1) inclusion of eye movement and 2) visual obstacle avoidance. This work is conducted using Santos™, a real-time predictive physics-based virtual human with a high number of degrees-of-freedom. With optimization-based posture prediction, joint angles serve as design variables used to minimize various human performance measures that provide objective functions, subject to constraints that represent biomechanical limitations and task characteristics. Vision-based objective functions and constraints are developed and easily implemented in order to accurately predict postures. First, two new degrees of freedom were added to the Santos™ model, representing vertical and horizontal movement of the eyes. Then, functions for eye movement relative to the head and body were developed based on experimental data. The new vision-based objective function expanded on the current vision model by incorporating these new functions. Additionally, a vision-based obstacle avoidance constraint was added in order to predict postures that incorporate the tendency to look around obstacles that may be in one’s line of site. Although vision alone does not govern one’s posture, when combined with other performance measures, more realistic predicted postures incorporating vision were obtained. Initial subjective validation suggests the predicted postures are accurate and realistic. The consequent capabilities have proven extremely useful for ergonomic studies and analyses of automotive cab scenarios.

Rights

Copyright © 2010 Lindsey A Knake

Share

COinS
 
Mar 27th, 12:00 AM

New Capabilities for Vison-based Posture Prediction

Although field of view (FOV) is a commonly used evaluation parameter with digital human models, minimal research has involved modeling how eye motion (relative to the head and body) affects the FOV and posture of a digital human striving to see a particular target. Few models incorporate independent eye movement and the effects of obstacles, with the ability to predict human posture realistically. This work presents two new and critical components for simulating how vision affects human posture: 1) inclusion of eye movement and 2) visual obstacle avoidance. This work is conducted using Santos™, a real-time predictive physics-based virtual human with a high number of degrees-of-freedom. With optimization-based posture prediction, joint angles serve as design variables used to minimize various human performance measures that provide objective functions, subject to constraints that represent biomechanical limitations and task characteristics. Vision-based objective functions and constraints are developed and easily implemented in order to accurately predict postures. First, two new degrees of freedom were added to the Santos™ model, representing vertical and horizontal movement of the eyes. Then, functions for eye movement relative to the head and body were developed based on experimental data. The new vision-based objective function expanded on the current vision model by incorporating these new functions. Additionally, a vision-based obstacle avoidance constraint was added in order to predict postures that incorporate the tendency to look around obstacles that may be in one’s line of site. Although vision alone does not govern one’s posture, when combined with other performance measures, more realistic predicted postures incorporating vision were obtained. Initial subjective validation suggests the predicted postures are accurate and realistic. The consequent capabilities have proven extremely useful for ergonomic studies and analyses of automotive cab scenarios.