Mobile Robot Object Recognition And Localization Based On Monocular Vision

Object recognition and localization is one of the basic tasks for mobile robots. It is also an important performance for the perception ability and intelligence of robots. Vision sensors can supply rich information about scene and have become the basic device in mobile robots. It is significant to do some research about vision-based object recognition and localization for mobile robot. Comparing with the panoramic vision and binocular vision, monocular vision has several advantages, such as simple structure, flexible motion and easy to calibration. The thesis focuses on monocular-based mobile robot object recognition and localization. The main work of this thesis is as follows:Firstly, the thesis introduces the background and significance about monocular-based object recognition and localization for mobile robot and reviews the state-of-art of vision-based object recognition and localization. The framework of the monocular-based object recognition and localization algorithm is given.Secondly, SURF algorithm is applied to extract features of object. BBF algorithm is used to match the features. RANSAC algorithm is used to eliminate the outliers. These methods can satisfy the real-time demand for object recognition and localization of mobile robots. The object recognition algorithm is realized. Experiments results show that the algorithm can obtain a good performance.Thirdly, the purpose of camera calibration and several common calibration methods are introduced. The simple Z. Zheng method is realized to calibrate our camera.Fourthly, after the task of object recognition is finished, the next motion of mobile robot is controlled according to the object position in the present image. And another image is obtained again after motion. The object features extracted from the two sequential images which are taken at different position are matched. The Structure From Motion algorithm is applied to get the relative position between robot and object combined with robot's encoder information. The object's location in the global coordinate is achieved based on the current position of robot.Finally, conclusions are given with recommendation for future work.