A review of Visual-Based Localization

Abstract

The visual-based localization (VBL) obtains the corresponding pose estimation in the localization system by utilizing various useful information in the surrounding environment, such as images, point cloud models, geometric information, semantic information. In recent years, visual-based localization (VBL) has been widely concerned by scientists, mainly because the commonly used GPS localization system cannot be effectively used in various environments. When GPS localization fails in some scenes such as very messy environments and severe signal occlusion, we can consider using visual-based localization to obtain the pose of the query images. Visual-based localization (VBL) has been widely used in the field of visual tasks, such as augmented reality, unmanned vehicle navigation, robotics, closed-loop detection, SFM (Structure from Motion) models. After years of development, the methods of visual-based localization (VBL) have been enriched and developed, In order to better understand the latest developments in VBL, overall research status and possible future development trends, we need make a systematic detailed classification of VBL. Although the predecessors have summarized the methods of VBL, due to the many new breakthroughs in VBL in recent years, the original summary is not perfect enough. So this paper will make a new and more detailed review of VBL in recent years. This paper divides the visual-based localization methods into three categories: image-based localization, localization based on learning model and localization based on 3D structure. And we also detail the principle, development of methods and the advantages and disadvantages of each method and future development trends.