A Ceiling Feature-based Vision Control System for a Service Robot

In this paper, we proposed a vision control system for a service robot which can provide navigation and assistance for daily activities. Compared with other sensors, vision sensor is cheap and practical, and can also perceive most of the information in the surrounding environment. Considering the variability of indoor objects, we choose an upward camera as the only sensor to extract objects and straight lines on the ceiling. The functional objects such as lights and alarms on the ceiling are detected by template matching to navigate the way of the robot. And the straight line features, which are rich on the ceiling, are used to correct the orientation of the robot. In our work, we clustered the angles of lines and refined the angles based on line lengths. The experiments have been conducted to confirm that straight line features detected in the image can calculate the orientations correctly and the orientations are more accurate with the refinement. In the final stage, we performed several tests on the system and the results demonstrate that the robot with vision control system has been successfully implemented the predesigned tasks.