Optimal Truncated MobileNet-Based Image Binarization for Pose-Based Visual Servoing of Autonomous Mobile Robot

Pose-based visual servoing (PBVS) can complement the frequent drift issue of light detection and ranging (LiDAR) coordinate of LiDAR-based simultaneous localization and mapping (SLAM), navigation, and servoing technology, especially when autonomous mobile robot (AMR) works on the automatic docking alignment missions for automatic pallet engaging or automatic battery charging whose alignment precision requirement is extremely stricter. But PBVS occasionally suffers from the poor detected image quality of ARTag landmark to cause the reading drift or error. This paper proposes a light-weight deep-learning image binarization method based on optimal truncated MobileNet model to preprocess the image quality of ARTag landmarks so that PBVS can evaluate the distance and pose between the ARTag landmark and the camera sensor more accurately, promptly, and steadily, for better feasibility of PBVS on the automatic docking alignment missions. Experimental results show, against conventional image-processing-based image binarization, conventional computer-vision-based image binarization, and conventional deep-learning-based image binarization, the proposed optimal truncated MobileNet-based image binarization not only raises the accuracy and reliability of ARTag’s reading, but also apparently improves the effectiveness and efficiency of PBVS’s operation, especially under environmental conditions of shadow occlusion, image blurring, low contrast, uneven illumination, or complex background.