A fast feature point extraction method for optical tracking system

Optical tracking systems (OTS) mainly consist of multiple cameras and optical target. A number of markers, at least three, are attached to the target. Position and motion posture of the target are obtained by photographing those markers. A small part of the image is occupied by marker. If markers are searched by traversing the whole image, huge expenditure and extravagant computing resources will be caused. A fast feature point extraction method is described in this paper to reduce computational burden. It combines marker prediction, perspective projection, nearest neighbor fast seed point search algorithm, region growing algorithm, and gray centroid method. Compared with the traditional whole image traversal method, the marker extraction speed is improved by hundreds or even thousands of times, which is verified by trinocular vision tracking experiment. The method is suitable for OTS.