As Unmanned Aerial Vehicles (UAVs) become increasingly available, pose estimation remains critical for navigation. Pose estimation is also useful for scene reconstruction in certain surveillance applications, such as surveillance in the event of a natural disaster. This thesis presents a Direct Sparse Visual-Inertial Odometry with Loop Closure (VIL-DSO) algorithm design as a pose estimation solution, combining several existing algorithms to fuse inertial and visual information to improve pose estimation and provide metric scale, as initially implemented in Direct Sparse Odometry (DSO) and Direct Sparse Visual-Inertial Odometry (VI-DSO). VIL-DSO utilizes the point selection and loop closure method of the Direct Sparse Odometry with Loop Closure (LDSO) approach. This point selection method improves repeatability by calculating the Shi-Tomasi score to favor corners as point candidates and allows for generating matches for loop closure between keyframes. The proposed VIL-DSO then uses the Kabsch-Umeyama algorithm to reduce the effects of scale-drift caused by loop closure. The proposed VIL-DSO algorithm is composed of three main threads for computing:​ a coarse tracking thread to assist with keyframe selection and initial pose estimation, a local window optimization thread to fuse Inertial Measurement Unit (IMU) information and visual information to pose scale and pose estimate, and a global optimization thread to identify loop closure and improve pose estimates. The loop closure thread also includes the modification to mitigate scale-drift using the Kabsch-Umeyama algorithm. The trajectory analysis of the estimates yields that the loop closure improves the pose estimation, but causes to scale estimate to drift. The scale-drift mitigation method successfully improves the scale estimate after loop closure. However, the estimation error level struggles to exceed the other state-of-the-art methods, namely VI-DSO and VI-ORB SLAM. The results were evaluated on the EuRoC MAV dataset, which contains fairly short sequences. VIL-DSO is expected to show more advantages when used on a longer dataset,where loop closure is more useful. Lastly, using the odometry as a feed, scene reconstruction and the effects of various factors regarding mapping are discussed, including the use of a monocular camera, camera angle and resolution in outdoor settings.