| 英文摘要 | For autonomous operating and service mobile robots, the purpose of localization is to determine the robot position reliably in real-time, which is necessary for autonomous navigation. Vision-based localization has many unsurpassable advantages, which is an important research topic in the robotic field. Several key techniques of vision-based localization for indoor mobile robots are studied in this thesis. The main contents of this thesis are listed as follows: Firstly, a new method is proposed to recognize the ceiling lines. According to the characteristics of ceiling lines, a new filter is designed to extract lines under the disturbances from lights and other objects. Different strategies are employed for line tracking, which makes the recognition of line features fast, robust and accurate. In addition, recognition methods of other objects, such as light, fan, and color mark, are discussed. Secondly, two visual self-localization methods are presented based on un-calibrated and calibrated camera, respectively. The former method estimates robot pose at adequate accuracy without knowing the camera intrinsic parameters when camera image plane is parallel to the ceiling; when the camera is calibrated, the latter method can compute robot pose without the parallel requirement. Thirdly, an approach of error correction is proposed for visual positioning. Natural landmarks in different areas of ceiling are utilized to correct accumulated errors of visual dead-reckoning method, which makes it practical for mobile robot to move in large and complex environment reliably. Fourthly, present a new method for robot to estimate intrinsic parameters by switching camera models. When lacking of the feature information, 1-parameter camera model is utilized. When having more information of features, 3-/4-parameter camera model is employed. This method ensures reliable estimation of camera intrinsic parameters, which enables the robot to approach to target at high accuracy. Fifthly, a visual method is presented to realize simultaneous localization and mapping. By the color characteristics of walls in the structured indoor environment, lines are extracted in real time to modify positioning results and create map simultaneously. Finally, summarize the whole thesis and discuss the future work. |
修改评论