Multifeatures Association Bundle Adjustment Constraints for Visual-Oriented Localization

Exploring various environmental features is beneficial to improve the environmental adaptability of the visual-oriented localization system. However, most existing visual-based localization systems mainly focus on extracting more kinds of landmarks from images but do not fully explore the geometric correlation among different features. This article proposes a multifeatures association bundle adjustment (BA) constraints method to improve the accuracy of the visual-oriented localization system. First, this article constructs an amount of spatial virtual plane (VP) based on point and line landmarks, which effectively riches the feature elements of environments. Second, a multifeatures association BA optimization method is proposed by exploring the geometric constraints of point-line-VP features. This optimization method does not need an additional plane extraction module and can realize the trajectory optimization of multifeature combinations of points, lines, and planes, which gives full play to the role of existing features. Finally, in order to balance all measurement residuals, a weight optimization method is proposed. The optimization module not only refines the position of landmarks but also obtains a more accurate pose of the device. The significant performance of the proposed method is demonstrated by comparing it with other state-of-the-art systems using public datasets and real-world experiments. Experimental results show that the trajectory estimated by this article can achieve more accurate and robust performance.