Swamidas, Joshua Arasakumar (1997) Predicting missing marker positions in simulated gait analysis systems. Masters thesis, Memorial University of Newfoundland.
PDF (Migrated (PDF/A Conversion) from original format: (application/pdf))
- Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.
Modern computer-aided vision motion systems provide a computerized and fully integrated tool kit for biomechanical measurement and analysis. These tools are useful for evaluation of problems, prescription of treatment and evaluation of such treatment. Many of these systems use reflective markers placed on key anatomical sites of the body to detect accurate three-dimensional spatial positions of the limbs being measured. While these systems ease automated data gathering, there are issues, such as the correspondence between an observed target and an established track, that require significant human intervention when markers disappear from view for short periods of time. When the system loses sight of a marker, it has no way of knowing where that marker will reappear and the track becomes broken or disjointed. Once the missing marker comes back into view, many current systems do not easily establish an association between the marker and its original track. -- In this thesis a solution to the problem of making correspondence between markers and their track histories was designed and tested. This solution also provided the capability of predicting the path of markers when they were out of view of the cameras. To test the algorithm three different repetitive motions were tracked using the Flock of Birds measurement system. -- The solution used a three-state Kalman filter to predict marker locations. The Kalman filter was coupled with constraints to determine matches between tracks and their corresponding marker positions. These constraints modelled a Region of Acceptance (ROA), distance from the center of the ROA to the last known position of a marker, and velocity matching. -- The Kalman predictor algorithm, because it is linear in nature, was able to predict the motion accurately while there was no change in acceleration. However, the Kalman predictor, coupled with the constraints, was useful in predicting and matching markers over a longer (100-500% longer) missing interval than the test case. -- To improve the prediction and matching capabilities of the Kalman predictor algorithm a physical motion modelm, that considers angular rotations at joints, was developed. The model is named the angular component model. This algorithm used an estimated or pre-measured motion model to check the location of the Kalman predictor. If the prediction did not match the model (within certain error bounds), it was corrected by the model algorithm using a calculation process that estimated the location of the marker based on its model. The addition of this algorithm to the Kalman prediction algorithm improved the prediction and matching capabilities. The matching worked well over the length of a 2 second gap (the longest used in testing) and the prediction of the marker path was excellent. The use of this model with the available tracking algorithms used in gait analysis will help in preventing the problem of disappearing markers in computer vision systems.
|Item Type:||Thesis (Masters)|
|Additional Information:||Bibliography: leaves 148-152.|
|Department(s):||Engineering and Applied Science, Faculty of|
|Library of Congress Subject Heading:||Gait in humans;Human locomotion;Human mechanics--Mathematical models;|
Actions (login required)