The first part of this project involves an investigation of current musculoskeletal modeling techniques and an attempt increase the accessibility and visibility of these techniques to clinical researchers. There are many software packages available for analyzing motion capture data, but a lot of researchers are forced to write their own software due to compatibility problems or prohibitive costs.
The top priority is determining the accuracy of data generated by various software packages, which will be done at UNC with Visual3D and MotionSoft, and at NC State with OpenSim (Figure 2), SIMM (Figure 3), and the AnyBody Modeling System. These tests can only be completed after creating software for data format conversion between common motion capture system formats and common motion analysis system formats. In the end, this will also help improve accessibility to these software packages and ease collaboration between laboratories.
Vico
n data at UNC has already been processed in Matlab and translated to formats that are easily usable by the SIMM and OpenSim systems. If OpenSim produces accurate results, it is desired to be able to import data into it, due to it being freely available and having an active development community. Current results seem promising for importing and analyzing motion capture data recorded by markers, but there have been greater errors in importing ground reaction force data from forceplates.
The second part of this project uses pattern recognition techniques to classify the conditions of patients performing tasks in a virtual reality environment. As a proof-of-concept, data that was recorded on some healthy patients and on some patients who have suffered a stroke will be automatically classified in order to identify if a patient has abnormal mechanics in their motion. Although this is a diagnosis that can be made fairly easily by a clinician, it is hoped that any developed techniques and feature spaces used for classification can later be expanded upon to help make more difficult diagnoses.
