[4L2-GS-13] AI application: Anomaly detection and maintenance

[4L2-GS-13-02] Detection of defective part of inside manhole using deep learning for automation of inspection.

〇Reon Katsumura¹, Takayuki Umeda², Shingo Andou², Jun Shimamura², Masaki Wada¹, Hiroki Shimabara¹, Takaaki Aihara¹ (1. Nippon Telegraph and Telephone East Corporation, 2. NTT Media Intelligence Laboratories, Nippon Telegraph and Telephone Corporation )

Keywords:infrastructure, object detection, CNN, deep learning, industry

Currently, we inspect annually about 30 thousand manholes within NTT East’s jurisdiction. We take pictures of inside manhole using 360-degree camera on-site. The repair judgement of manhole is carried out visually by many people at the centralized inspection center. Using Convolutional Neural Network, which has been successful in the field of image recognition, is expected to reduce work amount of the visual check with automation of the repair judgment for manhole inspection photograph. In this study, we automate detection of defective part of inside manhole using Mask-RCNN. And we verified the detection accuracy.