A new structure of spatial filters for real-time object tracking

Abstract

A simple and fast method of motion parameter estimation for tracking a moving rigid object is proposed. With the proposed method of spatial filter velocimetry, the spatial distribution of image motion or global motion parameters can be estimated by using a large number of small spatial filters arranged regularly in the image plane. Global motion parameters, such as translation, rotation, and dilation, can be isolated, identified and measured by evaluating the frequency distribution of the outputs from the spatial filters along a closed circular contour in the image plane. Experiments were performed in order to maintain the size and position of the object's image by using a camera of which the attitude is controllable. The proposed system, composed of only a commercially available PC, exhibited good tracking performances with processing rate of 15 frames per second.