6D object position estimation from 2D images: a literature review

Abstract The 6D pose estimation of an object from image is a central problem in many domains Computer Vision (CV) and researchers have struggled with this issue for several years. Traditional methods (1) leveraged on geometrical approaches, exploiting manually annotated local features, or (2) relied 2D representations different points view their comparisons the original image. two mentioned above are also known as Feature-based Template-based, respectively. With diffusion Deep Learning (DL), new Learning-based strategies been introduced to achieve estimation, improving traditional by involving Convolutional Neural Networks (CNN). This review analyzed techniques belonging research fields classified them into three main categories: Template-based methods, Learning-Based methods. In recent years, mainly focused which allow training neural network tailored specific task. For reason, most belong category, they turn sub-categories: Bounding box prediction Perspective-n-Point (PnP) algorithm-based Classification-based Regression-based aims provide general overview latest recovery underline pros cons highlight best-performing each group. goal supply readers helpful guidelines implementation performing applications even under challenging circumstances such auto-occlusions, symmetries, occlusions between multiple objects, bad lighting conditions.