Future robotic planetary exploration missions such as sample retrieval and in-situ resource utilization will require more accurate localization techniques such as Simultaneous Localization and Mapping (SLAM) to achieve the science goals. In this thesis, a visual SLAM system aimed towards computationally constrained systems is presented. In this work, Binary Robust Invariant Scalable Keypoints (BRISK) and Oriented FAST and Rotated BRIEF (ORB) feature descriptors are introduced and compared against Speeded Up Robust Features (SURF). BRISK is shown to achieve similar relative pose estimation performance than SURF while being an order of magnitude faster. This work also discusses a simple back-end pose-graph optimization approach using libg2o. The back end system improved the position estimation as well as detected loop closure events. These initial results show that computationally inexpensive feature detectors such as BRISK and ORB can be used as core feature detection algorithms for a visual SLAM system.