Method And Research On Apron Object Detection Based On Context Analysis

The Airport Collaborative Decision-Making(A-CDM)system is very important to improve the flight regularity.The traditional methods of A-CDM system data collection have problems such as high labor cost,low work efficiency and poor data reference.By intelligently detecting ground support vehicles in the apron surveillance videos,it realizes the automatic data collection of ground support operation time nodes of A-CDM system.This "smart airport" solution has attracted widespread attention from researchers.The technical core of this solution is the apron object detection method.However,there are still a series of difficulties and challenges in the actual object detection task in the apron scenes: the scales of objects in the apron are diverse,and the detection effect of some objects is unsatisfactory;in addition,due to the large difference in appearance within the interclass samples and the small difference between some interclass samples,general detectors that only rely on visual appearance features often have false detections.In response to these difficulties and challenges,a research on the apron object detection method based on context analysis is carried out,so as to improve the detection accuracy.Firstly,aiming at the problem that the scales of objects are in large difference and some categories of objects in the apron scenes are difficult to detect,a method of apron object detection based on multi-scale context enhanced feature is proposed.The scale context fusion method between layers is designed,which realizes the complementary expression of high-level semantic features and low-level detailed features;the multi-scale enhanced model is designed within the layer to obtain context information of different scales around the objects;combined with deformable convolution,the model can adaptively learn the deformation information of the occluded vehicle.Extensive experiments were carried out on the apron dataset,and the experimental results verify the effectiveness of the model.Secondly,the existing object detection frameworks only focus on visual appearance features.Consequently,it cannot take full advantage of the spatial context information provided by ground support vehicles in the apron.In view of the disadvantage,combining the business rules of the apron scene,an apron object detection method integrating spatial context-aware features is proposed.The graph convolutional network is used to encode the spatial geometric relationship of the apron objects,generating spatial-aware and regional-level features,so as to assist the object detection task;considering the redundant information that may be introduced in the process of context fusion,the attention mechanism is further combined to optimize the object features.Finally,in the experimental part,the two improved methods based on multiscale context and spatial context are integrated.The experimental results on the apron dataset show that the proposed method of apron object detection based on context analysis can significantly improve the detection accuracy.