A vision-based strategy for autonomous lift operation

Efficiency and accuracy are two crucial factors for autonomous robot operation. In this paper, we present a vision-based strategy for autonomous lift operation including panel and button detection method using a webcam. Two key problems are (1) scale variance during lift operation and (2) image deformation caused by the angle between robot and target. The approach presented here demonstrates its capability of handling these problems. The process has three steps: (1) panel hunting for a coarse localization; (2) panel verification at a close distance; and (3) button detection and localization. Our method shows the weak perspective model outperforms the affine model.