Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 28, 2023 - July 01, 2023
AR markers are utilized as a support tool for robotic object handling to easily estimate position and posture of objects with a monocular camera. However, the use of small-sized AR markers in a conventional camera system with fixed parameters causes a trade-off issue between the recognition accuracy and range. In this work, a precise object-handling system is developed using small-sized, high-resolution AR markers with a novel dynamic camera parameter control to realize both accuracy and wide-range of the recognition. The proposed method iterates calculation of camera parameters based on the measured depth distance of the AR marker. This camera control system achieves a recognition error of less than ± 1% over the entire depth distance of 1.0 m. Furthermore, the manipulation accuracy and recognition speed of the developed object-handling system were investigated with a two-step recognition strategy. During 220 trials of an object picking in total, a task success rate of 100 % and an average recognition time of within 0.5 seconds were achieved.
