Simulation and Experiment on Artificial Landmark-Based Monocular Visual Navigation System for Mobile Robot

Abstract

With regard to the automatic navigation of mobile robots working in a relatively fixed scene, this paper studies an artificial landmark-based monocular visual navigation system for mobile robot. The artificial landmark is designed to be conical in order to adapt to the rotation and distortion arising from the change of the angle of view during the movement of the robot. To better divide the artificial signpost, the inverse value of the H-channel mean of the background image has been calculated is selected as the main hue of the artificial landmark. The vision system adopts the template matching to detect and track the artificial landmarks that are previously installed in the working field within the field of view to determine the initial distance and heading. When the robot is approaching an artificial landmark, the node number of this landmark in the topological map is identified by detecting and decoding the Quick Response (QR) code pasted on the landmark, then the distance and position of the robot and the QR code plane are measured by using the monocular distance measurement technology based on an arbitrary plane constraint. The experimental results show that the detection of artificial landmark reaches 95% accuracy rate and averagely consumes 0.12 ms regardless of the changes in illumination intensity, angle of view, and dimension. The average root-mean-square error (RMSE) of the measurement by using the algorithm of monocular distance measuring is 10.9 mm, and the relative average error is 0.45%. In a word, the monocular visual navigation system proposed in this paper has a simple structure, fast detection speed, high precision and practical value.