Simultaneous Localization and Mapping (SLAM) is a process of sensing, estimating self-location and state, and charting a map of the unknown surroundings. It guarantees robots the ability to carry on more and more complicated tasks. Compared with other sensors for robots, camera is more economical, smaller-sized, easier to use and more informative. Besides, using a single camera simplified the whole SLAM research system with the same theory framework as those using more cameras.This paper aims to research and practise vision-based SLAM theories, mainly by analyzing visual feature processing problems in related work, and finally presenting a more robust visual feature operator for the test scene where viewpoint changes greatly in the image series. The presented method enhances landmark feature tracking accuracy and stability, which therefore improve the SLAM performance. |