Design of video based high level functions for a quadrotor type unmanned aerial vehicle

Among the video processing applications, aerial video surveillance is one of the interesting topics for researchers. Object segmentation and recognition system is one important topic for surveillance, and the other one is object tracking. There are various solutions developed with lots of researches to handle these problems. Wavelet transform analysis is one of the tools used to implement such applications because it yields useful information on image processing applications. In this work, two related application is implemented by taking advantage of wavelet transform. Gabor wavelets which provide analysis with different scales and orientations are utilized in both implementations. The aim of the first work is segmentation of the object from background image with the aid of Gabor wavelets, and classification of the pattern whether it belongs to the desired object or not. Nearest neighbor, k-nearest neighbor, Bayesian and artificial neural network classifiers are used to analyze the performance of the statistical features of the patterns. Bayesian classifier results in best accuracy. Artificial neural network also has a satisfactory performance to be used for classification. The other application is the tracking of an object on an aerial video taken from a moving observer. Tracking of the object is accomplished by using double window gradient based Gabor representation of a gray scale image and the extraction of pseudo-Zernike moment for feature selection to match best similar template around next expected pixel among consecutive frames. In addition, template matching process takes advantage of the weighted distance to find best matching template around an expected search point to obtain a better performance on tracking result. The algorithm to track the object on an aerial video satisfies the tracking task even under the circumstances where the observer is moving, the target object is moving or not moving, video record is low quality with noise, and there are high illumination changes.