抄録
In this paper, an algorithm to measure the sizes and the positions of some overlapping circular objects is described. The following algorithm is simplified by dealing with the binary image of circular objects. First, the boundaries of objects on the binary image are divided by the concavity points into some convex parts. The boundary pixels of each convex part are sampled at a constant distance due to the curvature of the convex part so that many sets of the three pixels which can define one circle are obtained. All sets of the sampled pixels belonging to the same circular objects are clustered and merged. Finally, the radii and centers of the circular objects are calculated by the minimum-squared-error-fitting method using the coordinates of the merged boundary pixels.
Furthermore, this paper describes the cafeteria automatic accounting apparatus as an application and discusses the experimental results. This method has a merit, i.e. the mean time of the measurement using by a 16-bit microprocessor (8086) is about one second per object. Also, this paper clarified the relationship between the accuracy of the measurement and the central angle of the convex part.
