1. Introduction
"Gait Kinematics Prediction Toolbox" predicts an arbitrary subject's gait kinematics using the body parameters of the subject. The toolbox is based on "KIST Human Gait Data" which contains 108 healthy human subjects' gait kinematics data, and Gaussian process regression (GPR) algorithm.
You can see the journal article related with this work in [1].
2. Download and Quick Start
One simple way to understand the toolbox is just to run the program in Matlab/Octave. You can download `Gait Kinematics Prediction Toolbox' and `KIST Human Gait Pattern Data' from the below links.
After the download, you can see and understand how to use `Gait Kinematics Prediction Toolbox' by running the script, demo_Gait_Pred.m in Matlab/Octave. It will produce several figures showing a test subject's predicted gait kinematics which consists of 14 joint motions and gait period. After some execution time (more than five minutes), you may see figures like the below.
Fig. 1: Sample outputs of toolbox
3. Description
The toolbox and database provide a function, Gait_Pred.m, mapping from an arbitrary subject's body parameters to the gait kinematics pattern with a statistical method. A user can predict an arbitrary subject's gait kinematics by only inputting the subject's 14 body parameters which are easy to measure. The output is a probability distribution (Gaussian distribution) of the predicted gait kinematics, not a strict value, of 14 joint motions and gait period. The mean value of the predicted probability distribution provides the gait kinematics trajectory, and the standard deviation value provides the level of confidence for the prediction. We provide a demo script, demo_Gait_Pred.m, for this function, Gait_Pred.m.
For user's special applications and for the expansion of the toolbox and database, we also provide a function, Gait_Pred.m, which creates a new mapping model. The user can make their own prediction model for the user's new or expanded database. For the format of the database, refer the manual in the toolbox. Optimization of the model may need considerable length of time (more than 6 hours) for optimization. We provide a demo script, demo_Gait_Pred.m, for this function, Gait_Pred.m.
For user's convenience, all the programs were made to be compatible with Matlab 7.x and GNU Octave. The main toolbox `Gait Kinematics Prediction Toolbox' is using another toolbox `GAUSSIAN PROCESS REGRESSION AND CLASSIFICATION Toolbox' for the GPR algorithm [3].
For more detail information associated with the theories and the prediction procedures, refer to [1] and the manual in toolbox, or contact Youngmok Yun (yunyoungmok |at| utexas |dot| edu).
4. Possible Applications
The range of its possible applications is broad from biomechanics analysis to rehabilitation robot control. Here, we briefly show an example analysis. The results were obtained from the given toolbox. The example shows the change of pelvis rotation angle trajectory with respect to age. For the prediction, we used the body parameters of the Table I.
Table I: Body Parameter Set for Prediction
Fig. 2: Prediction of Pelvis Rotation Trajectory with Respect to Age
Fig. 2 shows the relation between pelvis rotation trajectory and age of people. One possible analysis is that the range of motion of pelvis rotation decreases as getting old, and similar results are shown in [4].
5. Copyright
-Gait Kinematics Prediction Toolbox
Gait Kinematics Prediction Toolbox version 1.02 for GNU Octave and Matlab 7.x
Copyright (c) 2011-2014 Youngmok Yun (yunyoungmok@gmail.com). All rights reserved.
'Gait Kinematics Prediction Toolbox' is under freeBSD licence.
-KIST Human Gait Pattern Data
All copyright associated with `KIST Human Gait Pattern Data' is reserved, and the Center of Bionics at Korea Institute of Science and Technology has the right. This database can be used only for `Gait Kinematics Prediction Toolbox.' All redistribution and uses for other purposes in source and binary forms, with or without modification, are not permitted. KIST plans to open the database to the public completely. However, at this point we are making only the software publicly available. If someone wants to use the database for other purposes, please contact Youngmok Yun (yunyoungmok |at| utexas |dot| edu) or Changhwan Kim (ckim |at| kist |dot| re |dot| kr).
-GAUSSIAN PROCESS REGRESSION AND CLASSIFICATION Toolbox
6. Reference
[1] Y. Yun, H.-C. Kim, S. Y. Shin, J. Lee, A. D. Deshpande, and C. Kim, “Statistical method for prediction of gait kinematics with Gaussian process regression,” Journal of Biomechanics, vol. 47, no. 1, pp. 186–192, Jan. 2014.
[2] C. E. Rasmussen and C. K. Williams, “Gaussian Processes for Machine Learning,” The MIT Press, 2006.
[3] C. E. Rasmussen, "GAUSSIAN PROCESS REGRESSION AND CLASSIFICATION Toolbox version 3.2," http://gaussianprocess.org/gpml/code
[4] J. O. JudgeRoy, B. Davis, and S. Õunpuu, “Step Length Reductions in Advanced Age: The Role of Ankle and Hip Kinetics,” J Gerontol A Biol Sci Med Sci, vol. 51A, no. 6, pp. M303–M312, Nov. 1996.