Abstract:
Unmanned Aerial Vehicle has always seen a boom in its market since it was first introduced. From the first time when it was majorly introduced for military to now when it is even used by hobbyists in their recreational activity, the “un manned” vehicle always had a human operating it in its back-end. Automating UAVs is a great way to reduce human error chances and human fatigue in operation. Taxi devising and modeling of UAVs is an area in which complex equations are involved and such methodology in which research is still under progress, but adapting computer vision and sensor based automation approach is a feasible approach. We propose a novel technique that automates the taxi of the UAV. The proposed objective is achieved by correcting the UAVs trajectory on ground by measuring orientation data from IMU and Visual sensor mounted on the UAV. The data is processed using Machine Vision algorithms that include feature-based algorithms. The feature-based algorithms are mainly SURF, FAST, BRISK, HARRIS, MINEIGEN and MSER. The feature algorithms were all initially tested on single frames and then were approved for experimentation. The MSAC was used to remove outliers and find transformation matrix in between frames from the matching points. From these the orientation of the UAV is extracted by using geometric transformation technique and data was synced with the IMU readings. These readings are then fed into KALMAN filter that gives a corrected trajectory and a Root Mean Square Error for comparison of the Feature algorithms. Time taken, Feature Extracted, Feature matched, SNR, and RMSE comparison are all tabulated and briefly explained. In this thesis, we have observed that MSER has the lowest RMSE, SNR and the lowest time consumption. The experiments were carried 30 times and the data set was not taken from any other source except from our own experimental UAV runs.
