Proposal for a convergence criterion to the active net in two steps

In this paper, we propose a method for detecting a road sign from a scene image in the daytime. This method uses an active net in two steps. The active net is a deformable lattice network model which minimizes an energy function to detect the region of a target. In the proposed method, the first active net roughly estimates a position of a target. For the better target detection, we put the second active net around the estimated position. Its experimental results show the effectiveness and efficiency of the proposed algorithm. However, the active net can not occasionally detect the target region. This case shows that the progress of energy cannot converge smoothly. So we have developed the convergence criterion whether a good detection or an unsuccessful detection. That is, by evaluating features of the progress of the energy, the computation of the active net is automatically and fast stopped.