Object detection from UAV images with deep learning

Today, images taken from unmanned aerial vehicles are widely used in military, commercial fields, agriculture, security, and many more fields. In recent years, with the development of artificial intelligence technologies, object detection in images can be done accurately and quickly. The most frequently detected object types in unmanned aerial vehicles are moving objects and vehicle images. Vehicle detection is widely used in tracking and learning the coordinates of the target detected in military operations, in determining the density of vehicles in an open car park and directing the vehicles to empty parking places, and in calculating the traffic density at an intersection. It has shown that successful results can be obtained in object detection with popular models, such as Faster R-CNN, Mask R-CNN in Tensorflow library developed by Google. Models created on the basis of convolutional neural networks for object detection, firstly estimate the object in the test phase after training period, and then detect the object by enclosing the specified object in a frame. However, more advanced models let the image to be separated from the background texture by coloring the detected object. Mask R-CNN, thanks to its sample segmentation feature, differs from other algorithms by coloring multiple images with different colors, belonging to a single category in an image. On the other hand, the YOLO structure stands out with its speed in object detection by applying a single neural network to the image. In this study, vehicle detection has been done with YOLO and Mask R-CNN models, which are deep learning models, and it is aimed to reveal the most effective solution by subjecting the mentioned solutions to benchmark tests on sample data. In the training of the models, images were taken from different heights and locations using an unmanned aerial vehicle.