Autonomous selection, registration, and recognition of objects for visual SLAM in indoor environments

Abstract

SLAM is very important in autonomous navigation of a mobile robot. Mapping is the task of modeling the robot's environment and localization is the process of determining its position and orientation with respect to the global map. For successful SLAM performance, landmarks for pose estimation should be continuously observed. In this paper, autonomous recognition and registration of objects as visual landmarks is proposed for autonomous visual SLAM. SIFT and the contour detection algorithms are adopted to distinguish the objects from the background. Autonomous object recognition can enable the robot to recognize some objects without giving any object information to the robot and it can help the vision system to cope with unknown environments. Furthermore, by using object information, a small number of landmarks can be used in the same area compared to other visual SLAM schemes using corners and lines or scene recognition. Various experiments show that the proposed visual SLAM can improve autonomous navigation of a mobile robot.