Anomaly detection based on world coordinate features from a camera on a mobile robot

Product arrangement robots detect products to be arranged by using anomaly detection, because they are in anomaly state. However, general anomaly detection is mainly designed for a fixed camera, making it difficult to detect multiple kinds of products over a wide area while the robot is moving. Therefore, we focus on world coordinates of products. We have developed a new anomaly detection method that uses the high dimensional features obtained from 3-dimensional world coordinates. Experimental results showed that our method detected 87.0% of anomaly product, which was 12.6 points higher than representative methods for fixed cameras. Furthermore, a robot using our method successfully removed 85% of anomaly products from a shelf.