A Scene-Assisted Air-to-Ground Object Detection Method

Object detection based on deep learning has achieved outstanding performance in conventional field of view. However, under the air-to-ground field, coupled with the influence of background environment and camera shaking, the application of this kind of detection algorithm greatly decreases its performance, which poses a great risk for performing detection tasks on air platforms such as UAVs. This article discussed the difficulty of deep learning in air-ground detection and put forward a method of using Bayesian inference decision in the later stage of detection. Using the scene information to filter the detection results which made the detection process have the decision-making thinking greatly reduces the false detection rate.