Monocular RGB 6D object pose estimation for augmented reality: a survey

Accurate determination of the 6D pose of an object is important in Augmented Reality (AR) to align and anchor virtual elements within the real world. Achieving seamless and proper alignment of virtual objects within RGB image sequences enables AR to provide spatial value. For this purpose, the estimation of the 6D pose is one of the most relevant techniques, yet it remains a significant challenge. While there has been significant research in the field of 6D object estimation from RGB images, many challenges remain unresolved. Our analysis offers a thorough examination of modern methods based on deep learning due to their ability to deliver state-of-the-art results in 6D pose estimation, while addressing challenges such as changes in lighting, occlusions, background clutter and other environmental factors. We also consider the standard datasets and metrics to compare performance, perform a qualitative analysis from different perspectives and summarise the main technical challenges and trends in this topic, always from an application-oriented perspective in the context of AR.