Object Detection-Based Indoor Earthquake Risk Analysis for NAO Robot

In this study, the aim is to contribute detecting indoor objects with the NAO robot and classifying them according to the risk group they may pose during an earthquake, and thus to indirectly contribute to the potential use of the robot in earthquake awareness activities. Three separate data sets are collected for this purpose: a data set consisting of images taken with the NAO robot’s camera, a data set created from images collected in the Webots simulation environment, and a data set created from collected indoor images. Objects in the collected data are labeled in terms of three class categories as “safe”, “medium“ and “unsafe“ according to their potential life threading risk levels. Four different object detection models (YOLOv3-Tiny, YOLOv4- Tiny, YOLOv5 YOLOv8) are trained with this data and their performances are compared on sample images taken from all three data sets.