Robust estimation for camera homography by fuzzy RANSAC algorithm with reinforcement learning

In the computer vision approach, there are many problems of modeling to prevent affections of noises by sensing units such as cameras and projectors. In order to improve the performance of the modeling in the computer vision, it is necessary to develop a robust modeling technique for various functions. The RANSAC algorithm and LMedS algorithm have been widely applied for such issues. However, the performance is deteriorated when the ratio of noises increases. Moreover the computational time for the algorithms tends to increase for actual applications. In this study, a new fuzzy RANSAC algorithm based on the reinforcement learning concept is proposed for homography estimation. The performance of the algorithm is evaluated through experiments of camera homography. From the results, the method is found to be effective to improve calculation time, optimality of the model, and robustness in terms of modeling performance.