, Yongmin Shin

We discuss the current technology behind automatic selection of landmarks by simultaneous localization and mapping (SLAM), using a single camera in an unfamiliar indoor environment, and we propose an improved method. As currently implemented, automatic landmark selection by vision-based SLAM results in many useless landmarks, because features of the image are distinguished from the surrounding environment and are detected repeatedly. These useless landmarks create a serious problem for the SLAM system because they complicate data association. To solve this problem, we propose a method in which a robot initially collects landmarks through automatic detection while traversing the entire area where the robot performs SLAM and then, through clustering, selects only those landmarks that exhibit high rarity. This enhances system performance. Experimental results showed that this method of automatic landmark selection results in a high-rarity landmark being selected. Our method improves the performance of SLAM compared to conventional methods, an increases the accuracy of data associations