Moving Object Recognition for Airport Ground Surveillance Network

In this paper we first introduce an airport ground surveillance network, which is composed of data acquisition terminal based on multiple cameras, data transmission based on high-speed optical fiber, and processing terminal including some airport intelligent applications, e.g. intrusion warning and conflict prediction. Next we present a moving object recognition algorithm named AMORnet which is the basis of the intelligent applications in this surveillance network. Unlike the traditional object detection which cannot distinguish static and moving objects and moving object detection requiring accurate silhouette segmentation, the AMORnet only locate moving object and much faster than the time-consuming segmentation. To achieve this purpose, firstly we estimate the scene background through a motion estimation network, compared to the commonly used temporal histogram based approach, our background estimation method can better cope with infrequent aircraft movements in airports. Secondly, we use feature pyramids to perform regression and classification at multiple levels of feature abstractions. In this way, only moving objects are correctly recognized. Finally, experiments are conducted on an airport ground surveillance benchmark to verify the effectiveness of the proposed AMORnet.