Vision-based Localization Using by Line with Point for Feature loss

This paper presents a solution to overcome the shortcomings of point feature localization, and estimates the pose using points and lines. Vision-based localization estimates the position and orientation from camera information. It is used in unmanned driving systems, AR (Augmented Reality), and VR (Virtual Reality) to recognize the surrounding environment. Real-time performance is a challenging task for these systems to accomplish the missions based on the determined position. In this situation, ORB feature point extraction method is used. However, ORB feature extraction is based on points, position estimation cannot be performed if the feature cannot be extracted. In this paper, we handle this issue using our proposed architecture by extracting the line and using the number of missing features to estimate the camera's orientation. We also created an experimental environment in the actual world and confirmed the validity of the presented algorithm.