모바일 로봇으로 획득한 열화상 데이터의 이상감지

Abstract

In this paper, we propose an anomaly detection framework that can analyze thermal image data acquired by mobile inspection robots. The proposed framework includes image alignment algorithm (RANSAC-flow) to handle wobbling image data and extract time-series temperature data from the successive thermal images. Also, GAN-based anomaly detection algorithm is integrated in the framework to analyze the time-series temperature data and generate alarm. The proposed framework can be a general solution for the increasing demand on autonomous safety monitoring robots.