Visual SLAM in indoor environments using autonomous detection and registration of objects

摘要

For successful SLAM, landmarks for pose estimation should be continuously observed. This paper proposes autonomous detection of objects as visual landmarks for visual SLAM. Primitive features such as color and intensity, SIFT keypoints, and contour information are integrated to investigate environmental images and to distinguish objects from the background. Autonomous object detection can enable a robot to extract some objects without any prior information and it can help a vision system to cope with unknown environments. In addition, an adaptive weighting scheme and the use of a gradient of the gray scale are proposed to improve the performance of the proposed scheme. Using detected objects as landmarks, a robot can estimate its pose. A grid map of an unknown environment is built using an IR scanner and the detected objects are mapped in the grid map, which results in a hybrid grid/vision map. Visual SLAM using objects can have the less number of landmarks than other visual SLAM schemes using corners and lines. Various experiments show that the algorithm proposed in this paper can improve visual SLAM of a mobile robot.