HR-PD&GCNI: A Highly Robust Elevator Button Detection and Recognition Method based on Prior Distribution and GCN Inference for Autonomous Elevator Riding Robots

Abstract

Accurate detection and recognition of elevator buttons are essential for autonomous service robots. However, issues such as missed and false detections, caused by lighting, reflections, and damage, remain common in real-world environments. This paper presents a robust button detection and recognition method based on the prior button layout characteristics and graph convolutional network (GCN) inference. Using an improved YOLOv8n+SPD model, we cluster row and column coordinates to construct a layout map, predicting missed buttons, and improving recall. GCN is then applied to model spatial relationships and correct optical character recognition (OCR) false recognitions, enhancing recognition accuracy. Experimental results on public datasets demonstrate that our method achieves state-of-the-art performance, showing strong robustness in practical applications.