A Solution to Improve Object Detection for Images Captured by UAV-mounted Camera

Compare to traditional object detection, such as facial detection or general object recognition, the Unmanned Aerial Vehicle (UAV) object detection system is designed to detect and recognize targets with small scale from height. This paper proposes a solution combines neural network with multiscale feature maps and data augmentation to improve object detection for UAV using. The proposed network achieves 0.618 m AP and 16 fps on aerial image dataset.