Localization and navigation strategy play an important role in the realization of autonomous mobile robots. In this thesis, we focus on the vision based self-localization and navigation problem of wheeled mobile robots under unknown indoor environment. The main contributions of this thesis can be summarized as follows: Take into consideration that the texture characteristic of indoor environments is simple and there are many edge lines in the image, we proposed a stable feature detector through calculating the intersecting point of edge lines. Experimental results show that intersections are more robust to illumination changes than corners and blob features. A matching method of feature point set was proposed based on the estimation of homographs transformation. This method makes an assumption of homography relationship between the feature point sets, the matching result and the exact homography parameter can be achieved after repeated iteration. This method only involves the position of feature points, and has no relationship with the gray contract of images. Once this method was applied to the feature tracking, its tracking effect is much better than KLT tracking method, especially in inner ceiling images where the grey contract is homogenous. A stereo visual odometry is realized, and the motion parameters are estimated by stage to improve the location accuracy. The image sequence is segmented into several subsequences according to overlapped regions. The initial estimation of spatial point coordinates can be obtained through three-dimensional reconstruction, then the relative motion parameters of the subsequent frames are estimated with respect to the first frame, meanwhile the motion parameters and the locations of spatial points are optimized iteratively by the error of back-projection. This algorithm was realized using bundle adjustment method and achieves satisfied performance. A reactive fuzzy controller is designed for mobile robots navigation in unknown indoor environment. This controller follows the basic concept of artificial potential field method to build fuzzy rules for steering angle and linear velocity, and drives the robot to reach target along the local optimal path according to the reactive navigation strategy. During navigation, the robot can move along a smooth trajectory at a steadily changed speed due to its quick response to the obstacle changes. A virtual target switching approach is proposed to solve local trap situation prob...
修改评论