Object detection is one of the elementary problems in the field of computer vision and has a wide range of applications. This doctoral dissertation deals with the small object detection in aerial images gathered by an Unmanned aerial vehicles (UAV), colloquially known as drones. A specific example processed through this dissertation is person detection in aerial images for the purpose of developing a system that could be used in search and rescue operations. Namely, drones in search and rescue operations contributes significantly due to the ability to survey a larger geographical area from an aerial viewpoint. Collected images need to be visually inspected or processed automatically in order to find a person. Due to the height from which images are collected, objects (including people) are very small in those images, which further complicates the detection problem. In general, in such problems, the use of deep learning methods gives promising results, so through this research, different deep learning methods were used for the specific problem of person detection in aerial images. Through the dissertation, a new, multimodel approach that achieves better results than currently existing or proposed systems is proposed. In the second part of this research, the assumption was that the results could be further improved if the correlation of successive images collected during the search and rescue operation was taken into account, which was proven through this dissertation. Throughout the research, the HERIDAL image database was used, which was created for the purpose of detecting people and contains only the markings of people in aerial images, but not other objects. However, the proposed system can be generalized for application in the detection of other types of small objects, of course with a suitable image database for training and testing