Pose estimation from depth images

Pose estimation from depth images

Research Project
Project Goals

This project considers a model-based, Cartesian control theoretic approach for estimating human pose from a set of key features points (key-points) detected using depth images obtained from a time of flight imaging device. The key-points represent positions of anatomical landmarks, detected and tracked over time based on a probabilistic inferencing algorithm that is robust to partial occlusions and capable of resolving ambiguities in detection.  The detected key-points are subsequently used as input to a constrained, closed loop inverse kinematics algorithm which not only estimates the pose of the articulated human model, but also provides feedback to the key-point detection module in order to resolve ambiguities or to provide estimates of undetected key-points. Based on a standard kinematic and mesh model of a human, constraints such as joint limit avoidance, and self penetration avoidance are enforced within the closed loop inverse kinematics framework. We demonstrate the effectiveness of the algorithm with experimental results of upper-body and full body pose reconstruction from a small set of detected key-points. On average, the proposed algorithm runs at approximately 10 frames per second for the upper-body and 5 frames per second for whole body reconstruction on a standard 2.13 GHz

laptop PC.

Research Area

Publications

International Journal of Humanoid Robotics, 2015 2015
G. Bin Hammam, P. Wensing, B. Dariush, D. Orin
Automotive User Interfaces 2014 2014
K. Fujimura, L. Xu
Proc. of Digital Human Modeling Symposium, Tokyo, Japan 2014
C. John, B. Dariush
Proc. of International Summit on Human Simulation, St. Pete Beach, FL. 2012
B. Dariush
Journal of Computer Vision and Image Understanding, Vol. 114, Issue 12, pp. 1362-1375 2010
Y. Zhu, B. Dariush, K. Fujimura
International Conference on Robotics and Automation (ICRA). Anchorage, AK. 2010
G. Bin-Hammam, D. Orin, B. Dariush
Int. Journal of Humanoid Robotics Vol 6, No.2 2009
B. Dariush, M. Gienger, A. Arumbakkam, Y. Zhu, B. Jian, K. Fujimura, C. Goerick