| In the field of self-localization for autonomous mobile robot, cameras have lots of advantages over other sensors, therefore, the researches on visual localization for autonomous robot in unknown environment has become more and more important. However, there are disadvantages too, with the principal one being the high complexity of the visual localization algorithm, when robot roamed in the unknown environment, the increasing size of map database will leading to the problem of Positioning Low Efficiency. It will have adverse effects on the speed of localization, flexibility of robot, and memory space of robot. The dissertation focuses on optimizing the map database, streamlining the size of the map database without losing the localization accuracy.Using MRDS software simulation platform to build simulated Pioneer3DX and indoor apartment environment, then robot explored and mapped apartment environment autonomously. Extracted edge of images in database with robinson descriptor, and finally, got the initial, unoptimized map database.The dissertation discusses two map database optimization methods. The first one, matched the images in database with SIFT algorithm mutually, then measured the similarity between images by the number of matched points, optimized database by the way of measuring similarity; the second, according to the characteristics of path that robots roamed in unknown environment, deleted some related images which robot collected in straight path, optimized map database by the perspective of the geometric characteristics of apartment environment. At the same time, proposed an innovative distance estimation algorithm based on matches-coordinates, which offset the localization error caused by geometrical database-optimization. During the experiment of visual localization, the Dynamic Markov localization algorithm was used to analysis the accuracy of localization. Finally, compared with the original map database, the size of map database was optimized to79.1%. |
PDF Full Text Request