Landing Area Recognition by Image Applied to an Autonomous Control Landing of VTOL Aircraft

The pattern recognition aims to classify objects on different categories based on characteristics analysis. The usage of pattern recognition shows itself more and more frequent and widely used, covering different areas both in industry and research and development of new technologies. With that in mind, this work aims to compare two nonlinear classifiers, the Adaptive Boosting method and the Artificial Neural Network method, applied to the identification of a certain landmark, where the more profitable is inserted in a Vertical Take-Off and Landing (VTOL) aircraft real model to trigger the land action after a demanded mission in the trained pattern presence. It is used as sensing method, computer vision technique, from camera's acquired images the characteristics are extracted by a proceeding based on Viola-Jones technique. To optimize the classification, it is also used the Principal Component Analysis method to uncouple the amount of data in the training stage and optimize the results in both classifiers. To prove the efficiency of the classifier when the aircraft is flying, it is used to test a scenario where it is possible to simulate the landing action with different altitudes. The Adaptive Boosting method proved itself to be more advantageous due to its simple implementation and less computational processing effort, despite the slightly lower performance when it comes to classifying compared to the Artificial Neural Network. The Principal Component Analysis method also shows itself to be a good improvement when applied to both techniques, raising the success rate of the classifiers in all the tested cases. The results obtained in the simulation tests were considered satisfactory as the aircraft lands with great precision over the determined landmark after identifying the landing area used for training.