Implementation of target tracking methods on images taken from uav (Unmanned Aerial Vehicles)

In this thesis, a software has been developed by applying image processing, and deep learning methods to the images obtained with the camera on an unmanned aerial vehicle (UAV). Basic traffic objects are detected, placed in boundary boxes, labeled, and their confidence ratings are calculated as a percentage to provide an improved vision system for UAV users. The vision system has an important role in the professional field of UAVs to accomplish their tasks, and the target objects can be detected within the vision system by using Artificial Neural Networks (ANN). In this study, Faster-R CNN and YOLOv2 models were used, and compared for object detection. The Microsoft-COCO pre-trained data set is used for the training, and a new data set which is prepared with the images taken from UAV. Classes are person, car, and motorcycle. The objects were detected successfully. The methods and results were compared according to the performance and accuracy ratios both on the pre-trained data and on the data set prepared on the images taken from UAV.