A Novel Method for Real-Time Aircraft Detection in Airports

This paper presents a SSD (Single Shot multibox Detector) based aircraft detection method for airport video surveillance. The original SSD is not fast and accurate enough to meet the requirements of such real-time applications as airport surveillance. We introduce ResNet50 into SSD by using ResNet50 instead of the feature selection step in SSD framework. The proposed method and the original SSD are evaluated and compared based on a self-made aircraft database, which includes various cases as rainy, hazy weather, small targets, etc. Compared with original SSD, the proposed method is much faster, and the average accuracy is increased by 3%. Additional experiments based on public Pascal VOC database also indicate that the proposed method is better than the original SSD.