| With the rapid development of robot technology,human intelligence ability of the robots are becoming more and more stronger. In China, the development of robots has risen as a national strategies, the rise of service robot industry has become a trend. Although suppliers received many orders, the funding is not a lot, which caused by the industrial chain is not yet mature. The most fundamental reason is that which limited by technical difficulties, or its sensor is expensive, the underlying technology is not well developed. Therefore, in order to realize human intelligent household, more and more researchers put into service robotics.At present, the biggest problem of indoor mobile service robot is how to solve the navigation and localization. Localization is a very important function for the intelligent robots, the implementation method is also a lot. Track is simple and more mature, but it laid difficultly; Signal damping easily such as WLAN, Bluetooth, RFID and so on, which have poor accuracy; Ultrasonic is easy interference by obstacles, laser is too costly; The visual sensor processing’s speed is slow. But in many sensors, visual sensor can get rich information like human’s eyes, which is advantageous to the intelligent control, so localization based on visual research is more and more hot.Mobile robot based on the indoor environment as the research object, the paper mainly studies how to make use of the natural environment landmarks, namely, this paper solved the problem that the localization of the robots based on the characteristics of roof corner. In local feature extraction and description, the improved algorithm is proposed, and we use the improved algorithm for feature extraction and description for the image of the binocular vision, and complete corner identification and binocular matching range by the feature descriptor, finally, we finish the robot global localization by combining of feature point coordinate information using the trilateral measurement principle. In this paper, we mainly completed the following tasks through a large number of experiments:1)We analyze the research status of the localization system for indoor mobile robot, study the research background and significance, study the key technologies needed for robot localization, and finally get the method for the binocular vision, which was as a subject, this article, we study the theory research for the involved algorithm of binocular vision localization.2)We study the Open CV and its recompile technology, combine the SDK provided by Bumble Bee2 cameras for secondary development. According to compatibility issues from different development software and different versions of Open CV, through the COM component technology effectively solved the universality of interface. After the software establishing successfully, firstly, this paper completes the camera calibration experiment with the purpose of getting the parameters for localization and image correction.3)SIFT feature detection and description algorithm and FAST corner extraction algorithm are deeply studied. Combining the advantages of two algorithms, according to the actual environment, SIFT algorithm and FAST algorithm are improved, and we put forward the SIFT description algorithm for double threshold FAST corner detection, which greatly improving the accuracy and real-time performance.4) The paper introduces the sampling rules of visual landmarks, completes the relevant verification experiment. Aiming at the location problem of robot, the comprehensive experiments and research were carried on, and we further analyze the localization accuracy. Target tracking problem is discussed, we achieve the corner lock and tracking, further improve the real-time performance. Finally, we give the research plan of next step research.Finally, this thesis, depending on Open CV1.0+VC6.0 and Microsoft Visual Open CV2.49+VS2010 software, applies MFC modules and ALT technology to develop an application independently and realize the function of each module. It functions on the Traveler II Intelligent Robot platform successfully. |
PDF Full Text Request