Sub-Pixel object-Image Registration Using Improved Iterative Closest Point Method

In the non-structural industrial environment, accurate registration method against inaccuracy of clutter parts recognition and location by template matching method was investigated. First of all, it was proposed that the sub-pixel accuracy of pattern recognition was equivalent to solve image geometry transformation. Then, improved iterative closest point was illustrated, including using the dynamic neighborhood search strategy to locate matching points quickly, removing the false matching points based on the distribution law of matching points, using point to curve distance metric to construct the error metric function, obtaining the closed-form solution of error metric function. At last, simulation image and real image were tested accordingly. The results showed that the registration accuracy, positioning accuracy and real-time of the proposed algorithm were better than traditional method significantly. It met the requirements of subpixel registration accuracy and robustness in the unstructured environment.