To overcome these limitations, we propose an algorithm, called the CONTRACTING CURVE DENSITY (CCD) algorithm, for fitting parametric curves to image data. The method neither assumes object-specific color distributions, or specific edge profiles, nor does it need threshold parameters. To separate adjacent regions, we use local criteria that are based on local image statistics. We apply the method to the problem of localizing the ball and show that the CCD algorithm reliably localizes the ball even in the presence of heavily changing illumination, strong clutter, specularity, partial occlusion, and texture.

Hence, no training phase is needed. The local statistics define an with white lines; goals are blue and yellow; and expectation of "how the two sides of the curve robots are black with light blue or magenta might look." Second, refine the estimation of model parameters These stringent rules allow for simple mechanisms by (1) updating the mean of the estimation for object detection and recognition: in a maximum a posteriori step such that Segment the captured image into blobs of the the vicinity of the curve matches the expectation same color and interpret these blobs. To the defined by the local statistics and (2) updating best of our knowledge, all autonomous robot the covariance of the estimation based on soccer teams with vision-based perception apply the Hessian of the resulting objective function. However, because The two steps are repeated until there is no the RoboCup committee is planning to significant change in the estimated Gaussian make the rules more realistic, these objectrecognition distribution.