Vision based control of a quadrotor type unmanned aerial vehicle

Video based application development in images taken from unmanned aerial vehicles is preferred widely and problems encountered in the works considering surveillence and navigation made such topics a research focus. Because of the popularity of the applications of unmanned aerial vehicles, in this thesis, autonomous flight via a quadrotor is considered with flight in 3D virtual flight in Google Earth® virtual environment. As a first stage of the quadrotor navigation, classification of the objects seen by the vehicle. Background differencing and local thresholding methods are employed to distinguish the statistical features of the patterns and by the use of artificial neural networks and two different support vector machines, the target object is recognized. Based on the comparative tests, it is seen that the support vector machine based classifiers are seem to produce more reliable results than neural network based alternatives. In the sequel, quadrotor type unmanned aerial vehicle is tested in a bounded region in Google Earth® environment and the target is sought by using scale invariant feature transform technique. After determining the target area, the vehicle is commanded to surveil the area for a while then the vehicle lands to the area autonomously. As a last application, the vehicle is requested to synthesize the path in between two points in Google Earth® environment. Optical flow method is utilized and the necessary maneuvers are synthesized using the visual information. Results stipulate the high performance of the devised vision based autonomous flight system.