Tableware Recognition for Tidying-up Robot System

We determined the need to install an automatic tidying-up robot at the cafeteria's tableware return counter to save labor force, and we investigated tableware classification under complex conditions, which included seven types of tableware, leftovers, and garbage. We proposed a height-matching algorithm as a method to improve recognition accuracy in RGB-D image recognition. Two recognition methods were introduced: one based on contour recognition and the other based on the YOLOv7 algorithm, which was enhanced with height-matching. We validated the recognition capabilities of these two algorithms in complex scenarios. The contour recognition accuracy was 58.2%, whereas the height-matching-based YOLOv7 algorithm achieved 93.0%. Despite having significantly lower contour recognition accuracy than YOLOv7, it showed promising research potential in combating interference.