Monocular 3D Object Localization using 2D Estimates for Industrial Robot Vision System

Abstract

3D Object Localization has been emerging  as one of the main challenges in Machine Vision tasks. In this paper, we proposed a novel 3D object localization method, leveraging a blend of deep learning techniques primarily rooted in object detection, post-image processing, and pose estimation algorithms. Our approach involves 3D calibration methods tailored for low-cost industrial robotics systems, requiring only a single 2D image input. Initially, object detection is performed using the You Only Look Once (YOLO) model, followed by an R-CNN model for segmenting the object into two distinct parts, i.e., the top face and the remainder. Subsequently, the center of the top face is served as an initialization position, and being refined with a novel calibration algorithm. Experimental results demonstrate a notable reduction in localization error by 87.65% when compared to existing methodologies.