Research On Visual Localization Based On The Color Image

The mobile robot was first studied in early 1960s', and up to now, it has been successfully applied to military, civil and scientific fields. Of the research studies on the mobile robot, navigation is a basic and core technique. Since the vision sensor can capture and describe rich and overall environment information, visual navigation has already become a significant method of navigation for the mobile robot.With reference to visual navigation and localization for the mobile robot, camera calibration, color image processing and visual localization are studied in the thesis which includes the following six sections.Firstly, the research development of the mobile robot, visual navigation and localization for the mobile robot is briefly reviewed. Besides, the theory, requirement and realization of vision systems for the mobile robot are deeply analyzed.Secondly, the imaging relationship between the spatial point and its corresponding image point shall be first established. That's what camera calibration does research on, presenting imaging model and an algorithm of how to solve the internal and external parameters of the model. The theory of camera calibration is studied and two calibration methods, including the linear least-squares solution method and two-stage method, are analyzed in detail.Thirdly, the object detection and recognition based on color information is studied. Furthermore, a color image segmentation and recognition method based on HSI color model is presented. Converting the captured image from original RGB model to HSI model, it can segment the image and recognize the colored objects by threshold method whose thresholds depend on both H and I.Fourthly, visual localization and motion estimation in indoor environment is studied. A monocular vision system is presented to deduce a visual depth measurement model by means of landmarks, perspective projection imaging model and spatial geometrical relationship, thus localizing the mobile robot and a moving object. Then, a feature-based motion analysis method is proposed to estimate motion parameters of the object from sequential images. Based on the motion parameters, a motion model for the robot to intercept the moving object is deduced.Fifthly, camera calibration experiment based on Zhang's two-stage method is carried out with high precision. The proposed color image segmentation and recognition method is validated through the experiment on extracting objects of different geometrical shape. The proposed monocular visual localization method is validated through the experiment on localizing the yellow ball.Finally, research results of the thesis are summarized and future work is addressed.