Simultaneous Localization and Mapping (SLAM) is the key to implementing truly autonomous mobile robots. Landmark characterization, data association and computational complexity are vital in achieving a practical SLAM implementation. On the one hand, vision sensors can provide rich information about shape, color and texture, which is useful to form reliable data association and predict a closing loop; on the other hand, under the present research level and technical conditions, rational design of artificial landmarks is an effective means for implementing SLAM in a complex environment. In view of this, supported by the National Natural Scientific Foundation, this thesis has studied on the visual SLAM method based on a novel artificial landmark system. The novel work and contributions of this thesis includes: Firstly, an improved recognition algorithm is proposed for the false recognition problem during the rapid movement of the robot, which improves the recognition accuracy and contributes to solution of the data association problem in SLAM. Secondly, a practical error model for odometric position estimation is proposed on the basis of analysis and verification of the motion model and observation model. An improved extended Kalman filter (EKF) SLAM algorithm based on mixed data association is presented, which improves the localization precision of the robot and the map accuracy. Experimental results verify the effectiveness, robustness and consistency of the algorithm. Thirdly, an extended multiple fading strong tracking filter (EMF-STF) SLAM algorithm is proposed and the calculation method of the multiple fading factors is also described. The algorithm is able to decrease the error induced by the linearization, improve the localization precision and the map accuracy and constrain the covariance within a small range to enhance the credibility of the map. Fourthly, making use of the MR code, a practical topological navigation strategy is proposed for indoor mobile robots when the environment map is known, combined with an approaching control algorithm and an extended line tracking algorithm. Experimental results show the rationality and effectiveness of the strategy in the real environment. Finally, on the basis of the above work, a visual SLAM and navigation experimental system based on MR code has been implemented and an experimental software SLAM-NAV with a friendly man-machine interface has been developed. Utilizing the system, the robot can comp...
修改评论