The three-dimensional measurement of unconstrained motion using a model-matching method

Abstract

The measurement of motion is a basic technique for the quantitative mechanical analysis of human movement. However, previous methods of motion measurement using goniometry and surface markers have several associated problems: goniometers constrain human motion; and surface markers are missed, etc. To resolve these problems, a new technique of motion measurement using image processing is proposed. In this method, movements of the whole body are captured as image information, and a geometric model of a human body is fitted onto the contours of the image sequences. The joint angles are then estimated from the model. Usually, every segment has to be coloured separately so as to fit the geometric model automatically into the images; furthermore, this takes a very long time to process. Therefore, we reduced the processing time using the information in the overlap area between the models and images, instead of contour information. We estimated the joint angles of hidden segments with the grey scale image information. Thus there is no need to colour segments and attach surface markers; and unconstrained motion can be measured. Using the method developed, three-dimensional motion of the fingers including 21 segments was measured in 3–5 min per frame by a personal computer. The errors are 2° maximally.

Keywords:

• Image processing
• Finger movements
• Measurement of motion
• Model-based vision