Research On The Recognition And Tracking Problem Of Arbitrary Ball Based On Omnidirectional Vision For Soccer Robots

This thesis designs and realizes an arbitrary FIFA football recognition and tracking algorithm based on omnidirectional vision to solve the arbitrary FIFA football recognition problem in the RoboCup Middle Size League competition. The perception ability of the Middle Size League soccer robots can also be improved.Firstly, the thesis designs and realizes an arbitrary FIFA football recognition algorithm based on omnidirectional vision system, and the system deploys the improved Haar-like feature and AdaBoost learning algorithm. The recognition system includes two phases: the off-line training phase and the on-line recognition phase. In the off-line training phase, dozens of panoramic images which contain different kinds of arbitrary footballs are captured by the omnidirectional vision system of NuBot soccer robot. And these images are preprocessed and used as the training set. Then the Haar-like feature vectors are calculated with each image of the training set and are combined as a feature matrix. Finally, the AdaBoost learning algorithm generates a classifier with the feature matrix as the input. In the on-line recognition phase, a series of rectangle windows are defined along the radial direction of the panoramic image. According to the imaging characteristic of the omnidirectional vision system, the lengths of the rectangle windows vary along the radial direction of the panoramic image. Then the whole panoramic image is searched by these rectangle windows along both the orbicular and radial direction. Finally, the classifier is applied to classify these rectangle windows, which means the classifier judges whether the window contains a ball or not.Secondly, the RoboCup Middle Size League competition is a highly dynamic and competitive environment which requires the real time performance in the object recognition. To deal with that the proposed recognition method can not perform in real-time, this thesis designs a football velocity estimation algorithm based on Kalman filter and RANSAC to track the movement of the ball. Firstly, several ball positions are restored and filtered by the Kalman filter. Then RANSAC algorithm is used to calculate the velocity of the ball. In the simulation and practical experiments, the proposed algorithm performs better than the one based on Least Square Method and Kalman filter. The velocity can be calculated by the ball velocity estimation algorithm, the movement of the ball is also anticipated. The arbitrary ball recognition algorithm can search the ball on the base of the anticipated movement of the ball, so the number of search windows and the computation load decreases. The real-time performance of the arbitrary ball recognition algorithm can be improved greatly by integrating the ball velocity estimation algorithm.Finally, several ball recognition experiments are performed. The experimental results show that the recognition algorithm performs well, and can be run in real-time by combining the ball veclocity estimation algorithm. Furthermore, the ball can be recognized and tracked effectively even when the ball is occluded or the lighting condition changes. The proposed object recognition method based on the omnidirectional vision can also be applied in other generic object recognition problem by substituting the training set.