Detection method for transparent window cleaning device, image processing approach

Jiseok Lee,Hwa Soo Kim,Kyungmin Kim,Taewon Seo,Hobyeong Chae

doi:10.1038/s41598-022-07235-y

Jiseok Lee, Hwa Soo Kim + Show 3 more

Open Access

https://doi.org/10.1038/s41598-022-07235-y

Copy DOI

Journal: Scientific Reports	Publication Date: Feb 25, 2022
Citations: 1	License type: open-access

Affiliation: Hanyang University, Kyonggi University

Abstract

Recent years, there has been an increase in the number of high-rise buildings, and subsequently, the interest in external wall cleaning methods has similarly increased. While a number of exterior wall cleaning robots are being developed, a method to detect contaminants on the exterior walls is still required. The exteriors of most high-rise buildings today take the form of a window curtain-wall made of translucent glass. Detecting dust on translucent glass is a significant challenge. Here, we have attempted to overcome this challenge using image processing, inspired by the fact that people typically use just the ‘naked eye’ to recognize dust on windows. In this paper, we propose a method that detects dust through simple image processing techniques and estimates its density. This method only uses processing techniques that are not significantly restricted by global brightness and background, making it easily applicable in outdoor conditions. Dust separation was performed using a median filter, and dust density was estimated through a mean shift analysis technique. This dust detection method can perform dust separation and density estimation using only an image of the dust on a translucent window with blurry background.

Highlights

Background ignore scoreFor each edge detection algorithm, a test was conducted to determine how blurry the background needed to be before it could not be detected
A dust separation stage was performed through a median filter, and dust density estimation was performed through mean shift clustering
Experiments conducted with a test bench showed that the estimated dust density was directly proportional to the true dust density

Summary

Introduction

For each edge detection algorithm, a test was conducted to determine how blurry the background needed to be before it could not be detected. This depends on the camera, but the background must focused on to use edge detection. At this stage, the larger the minimum blur level that ignores the background, the larger the minimum distance between the transparent window and the background objects. The best method to use is a method that can separate the background and dust pixels with the least blur.

Methods

Conclusion