[POSTER] CamSLAM: Vision Aided Inertial Tracking and Mapping Framework for Large Scale AR Applications

In this paper, we introduce CamSLAM, a simultaneous localization and mapping (SLAM) framework composed of a powerful visualinertial odometry backbone using an error-state Extended Kalman Filter (EKF) for sensor fusion, and a very efficient and lightweight parallel mapping engine utilizing keyframe based pose graph data structure and binary descriptors for feature matching and indexing. The framework is capable of generating and maintaining large scale maps of several kilometers in length that are typically required for AR applications in military training sites and industrial applications in environments such as factory floors, office buildings or warehouses. Both the online map creation and refinement, and navigation based on a pre-loaded map is capable of running in real time on a mobile processor such as Nvidia Tegra X1.