Vision-based contamination detection algorithm for facade cleaning

Abstract

Recently, as many high-rise buildings have appeared, research on facade cleaning robots is increasing. The ability to detect contaminants in a cleaning robot has many potential applications. information of detection contaminants can adjust the cleaning parameters or used in an automatic exterior wall management system. This study aims to solve this problem with a vision-based method, focusing on the fact that humans detect contaminants on the outer wall with only naked eyes. Through this, the contaminant information of the facade is extracted by processing the image acquired through the camera. We design an independent module for each type of contaminant, and present a framework that processes them sequentially. To prove the effectiveness of this method, a test specimen and a test bench were designed. For the object-type and area-type contaminant of the outer wall, the minimum IoU was 0.6867 and 0.7526, respectively, and the Coefficient of Determination was 0.967 for the particle-type. A study on dust detection of translucent glass showed a coefficient of determination of 0.987