Study On Object Recognition And Ilcalization About NAO Robot

In RoboCup standard platform group, NAO perceived the field’s real-time information rely mainly on it’s camera, in order to adapt to the complex environment of the field, image processing algorithms have to be real-time and robustness. The part of image processing must extract the target form images rapidly, and this paper described the method of ball, goal and white line. Robot must complete the task of self-localization, localization is a key method and its meet to the level of robot’s intelligence. Monte Carlo Particle Filter Localization algorithm (MCL) is widely used in robot, many improved method improved the MCL algorithm. In this paper, the research platform is NAO robot, researched target recognition and MCL particle filter algorithm.Firstly, segmenting the image accord to the targets’color, in order to avoid the impact of outside information, a method of determining the field boundary is presented. According to the field boundary is non-concave, this paper presented a method though finding the maximum slope, determined the field’s boundary quickly and accurately, avoided the information of outside field and reduced the subsequent target recognition in image processing area.Secondly, on the basis of image preprocessing, started from the reality situation, achieved the recognition of ball and goal. The recognition of ball and goal is based on different color, and used Hough method when recognized the ball, in order to distinguish the goal post, we calculation the angular point.Thirdly, because of the NAO robot white shell, making the white line recognition is more difficult. This paper presented a new method, found the white edge though grid scanning at first, then traversed the white edge, classified the edge accord to the vary of slope, we can get the result after Hough Transform. This method is simple and effective, meet the requirement of real-time and robustness for NAO robot.Fourthly, MCL algorithm is an iterative algorithm based on probability calculations, solved the problem of non-Gaussian because of "kidnapping" perfectly, completed the robot localization effectively and increased its stability. In order to reduce the amount of computation, this paper introduced the method of partition of field, educed the MCL calculation of sample space, and improved the algorithm.In this paper, we considered the actual situation, simplified the method of target recognition, especially improved the recognition of white line efficiently, reduced the space of robot pose and introduced the MCL improved algorithm to speed up the localization algorithm. Finally, we summarized the work, and given with recommendation for future work.