Anomaly Detection using Elevation and Thermal Map for Security Robot

This paper is about an algorithm that displays the elevation and temperature of a surveillance area in the form of a map using a mobile robot platform equipped with a Lidar and a thermal imaging camera, and determines whether it is abnormal. Localization is performed using Lidar, and the point cloud and thermal distribution at the corresponding location are expressed as a top projection type map. The collected elevation and temperature distribution maps are checked for anomalies using the first-stage auto-encoder and second-stage CNN algorithm. We collected data on normal situations and abnormal situation DB produced by various size boxes and gas burners to study. As a result of the study, it was possible to detect 74.4% for the first stage, 94.5% for the second stage, and 80.2% for the first stage and 99.3% for the second stage for the thermal map. The results of this study will be used in areas such as security robots and guide robots.