DeepLab V2 기반 실내 시설물 손상 탐지 및 추출 알고리즘

Abstract

This paper proposes an algorithm that displays the damage of main indoor facilities by detecting and extracting information. First, to extract information for each facility structure, a facility segmentation algorithm is implemented by using DeepLab V2. For efficient learning of the network, the pretrained ResNet101 network is used for transfer learning. In addition, the damage extraction algorithm of the indoor facilities is also based on DeepLab V2, the same network structure used in facility segmentation. It is confirmed through a robot experiment that the accuracy of representing the damage information regardless of near and far facilities is greatly improved when the facility segmentation algorithm is used.