Abstract
In the field of robotics applications, sensing the environment is a crucial step, and recognizing highly reflective surfaces poses a significant challenge. Surfaces with high reflectivity, such as mirrors and glass, can significantly affect sensor data, including depth values from depth cameras and distance values from radar ranging. In this paper, we propose a method for real-time recognition and segmentation of specular surfaces in motion. We extract RGB images, depth data, and motion information captured by a depth camera during motion and utilize the SLAM camera principle to predict feature points. We analyze the displacement characteristics of feature points on reflective surfaces in motion images and subsequently design an algorithm for detecting anomalies in specular features using the DBSCAN clustering method. We integrate the detection of specular surface anomalies and boundary segmentation methods to obtain segmentation results for the specular surfaces in the image.
