Vehicle Detection in Aerial Images Using Modified YOLO

In recent years, object detectors have been developed significantly, many general object detectors have sprung up. However, these object detectors are design for general object, and do not directly apply well to detecting vehicles in aerial image. Vehicles in aerial image occupy smaller pixels comparing to general object in general data set, and have special angle. General object detectors are prone to miss detection and Confuse the roof of the car with the roof of the house in aerial image. Therefor we improved YOLOv3 in order to resolve the task of vehicles detecting. Increasing the depth of network is applied for enhancing network fitting, top-level feature maps are called to provide more detail information for getting better detection effect. Through a series of network structure changes, our algorithm performance excels State-of-the-art aerial vehicles detection algorithms.