Self-detection from robot-mounted camera image only by comparing multiple scenes

Abstract

We propose a method to distinguish the robot itself from the surrounding environment in the camera image mounted on the robot.In this research, we aim to detect robot itself by using only images of multiple scenes in which the robot itself and the different background are shown for comparison, and extracting similar areas by feature value. In this paper, we confirmed that the robot and background can be separated using the proposed method. Experimentally, feature points were detected with a plurality of feature values, and after matching them, a histogram was created. It was confirmed that the robot and the background can be separated roughly by setting the thresholds from this histogram.