Moving target detection and tracking on thermal and hyperspectral images

Digital imaging systems have become a necessity of daily life with the advances in the technology and being used in most areas. It has been catching attention on itself in some areas related to security, in particular military forces, due to problems stemming from target detection and tracking.Thermal imaging systems are commonly used in military applications because of their feature sensitive to heat which enables to view in nights; automatic target detection and tracking in different images.In addition to that, hyperspectral images (HSI) have a special place in the field from other image types due to being sensitive to the type of material on the surface of the image and providing high spectral resolution.In this thesis, automatic target detection and tracking has been realized on thermal images with the methods of SIFT Flow and Covariance Descriptor (CD). The suggested method is applied in different types of targets and tracking scenarios.For target detection and tracking on HSI, different types of superficial targets are placed over hyperspectral data set, and moving scenarios are superficially produced again. For motion detection, SIFT Flow; for detection in the mentioned area, CD methods are used. Furthermore, on HSI, target recognition is realized by putting different approach to CD method called partial-Covariance Descriptor (p-CD).In the experiments, the performance of the suggested way is evaluated in thermal images by confusion matrices. On HSG, the performance of recognition is compared with both classic CD approach and SVM. As a result, it has been found that the suggested way for target recognition gave better results than radial SVM in some classes.In this thesis, our aim is to develop an algorithm which can be usedfor security applications of thermal images under various scenarios.As for HSI, we aim to detect, and track motion of pixel leveltargets while using both spatial and spectral information. We conclude that proposed methods are convenient for this purposes.