The detection of foreground targets on airport surface is the foundation of airport surveillance applications, but there are no object detection datasets specifically designed for movable targets on the airport surface. AFTD is an application-oriented dataset for movable target surveillance on the airport surface. Currently, the AFTD dataset has a 10050 image data containing 25968 aircraft instances, 24388 vehicle instances, and 5745 person instances, which encompasses a variety of challenges on airport surface movable target detection, such as super multi-scale, multi-level occlusions and viewangle changes. We use manual annotation to label the images.
AFTD contains 10050 image data and below are the variations of movable target detection on airport surface.
Super multi-scale: Compared with the scale change in ordinary monitoring scenes, the scale of aircraft in airports varies greatly on the imaging plane. The airport has a vast area, with monitoring distances spanning the range of ten meters to kilometers. When the scale of the aircraft is too large, details may be magnified, resulting in the algorithm being unable to effectively capture the global features of the object. When the scale of the aircraft is too small, it is difficult to effectively capture the local features of the object.
Multi-level occlusion: In airport, there is a frequent occurrence of aircraft being obstructed by other objects, such as bridges, adjacent aircraft, etc. and it is also frequent that part of the aircraft's body is outside the screen. Occlusion often leads to incomplete feature information of the object, or even results in a serious lack of key identification features.
Viewangle changes: The appearance of the aircraft varies tremendously from one viewangle to another because of the wingspan shape. The shape of an aircraft is completely different in the side and front and rear viewangle.
Illumination changes: Airports, as a typical outdoor scene, make all the data in AFTD subject to illumination variations.
Weather changes: There are various weather changes in airports and different weather conditions have a significant impact on the visibility of the outdoor environment.
Complex background.: In addition to movable targets, there are a large number of buildings and greenery in and around the airport, which constitute a complex visual background.
The AFTD dataset provides precise bounding boxes for movable targets on the airport ground, and five students were involved in the labeling effort.