AVIS: An Innovative Image Preprocessing Method for Object Detection of Aerial Images

Recently, discussions on aerial communication platforms using drones, balloons, and high altitude platform stations (HAPS) have become widespread in the communication industry. Besides providing several aerial wireless communication services, aerial images can be applied in many scenarios, such as object detection. Investigating existing object detection models in aerial image processing shows that they have lesser precision than the ground use case. Therefore, we propose the aerial video image segmentation method (AVIS) to substantially improve the detection precision. The segmentation method is expatiated and evaluated. Furthermore, an additive targeted area masking method (TAM) was proposed to suppress the computation load. Compared to existing methods, the evaluation results show that our proposed method greatly improves the aerial object detection precision.