Fusion of IMU and Monocular-SLAM in a Loosely Coupled EKF

Sammanfattning: Camera based navigation is getting more and more popular and is the often the cornerstone in Augmented and Virtual Reality. However, navigation systems using camera are less accurate during fast movements and the systems are often resource intensive in terms of CPU and battery consumption. Also, the image processing algorithms introduce latencies in the systems, causing the information of the current position to be delayed. This thesis investigates if a camera and an IMU can be fused in a loosely coupled Extended Kalman Filter to reduce these problems. An IMU introduces unnoticeable latencies and the performance of the IMU is not affected by fast movements. For accurate tracking using an IMU it is important to estimate the bias correctly. Thus, a new method was used in a calibration step to see if it could improve the result. Also, a method to estimate the relative position and orientation between the camera and IMU is evaluated. The filter shows promising results estimating the orientation. The filter can estimate the orientation without latencies and can also offer accurate tracking during fast rotation when the camera is not able to estimate the orientation. However, the position is much harder and no performance gain could be seen. Some methods that are likely to improve the tracking are discussed and suggested as future work.