基于翼尖检测的航空器机位滑入过程冲突判定

Wingtip detection-based aircraft gate taxi-in-conflict determination

针对航空器滑入机位过程中可能出现的翼尖刮碰问题,设计基于翼尖关键点检测的航空器机位滑入过程冲突判定方法。采用机位摄像头获取目标数据,设计含目标检测、关键点检测、坐标转换及冲突预告警的航空器入位冲突检测模型。对传统HRNet(High-Resolution Net)进行精简,实现网络轻量化处理。考虑以往基于视频成像像素实施冲突检测时存在的较大误差,设计摄像头坐标转换算法,提高冲突判定的准确性。综合航空器机位内的静态间隔标准和动态避撞过程,提出翼尖安全区划设方法。针对入位过程中车辆侵入翼尖安全区的潜在刮碰情形,设计冲突检测模型,并构建某机场的机坪物理沙盘以实施模拟验证。结果显示,模型预警平均F1分数为92.22%,告警平均F1分数为97.93%。

As the most stressful area for airport traffic,apron operation safety is getting increasing attention.For the possible wingtip scraping problem in the process of aircraft taxiing into the gate,a conflict determination method for the aircraft gate taxi-in process based on wingtip keypoint detection is designed.The aircraft gate taxi-in conflict detection model with object detection,keypoint detection,coordinate conversion and conflict pre-warning is designed using the apron activity video as the monitoring data source;Combining the needs of aircraft wingtip detection in the apron,the fourth level branch and the fourth stage of the original HRNet are streamlined to obtain the HRNet lite network and obtain the lightweight processing of the network;In view of the shortcomings of the traditional video conflict detection based on the distance determination of imaging pixels,a coordinate conversion algorithm for fixed surveillance cameras is designed to convert the coordinates of imaging pixels to the actual coordinates located on the apron surface to improve the accuracy of conflict determination based on distance determination;Considering the static spacing criteria and dynamic collision avoidance process for aircraft activities in the apron,a method is proposed to delineate the wingtip safety zone containing the early warning zone and the warning zone,and a corresponding conflict detection model is designed for the scraping after the vehicle invades the wingtip safety zone;The physical sandbox of an airport corresponding to the apron is constructed to validate the aircraft wingtip conflict detection model,and the results show that the average F1 Score of model warning is 92.22%and the average F1 Score of early warning is 97.93%.The model is less affected by the changes in aircraft taxi-in speed,aircraft wingspan,and vehicle speed,which has strong robustness.This project can be extended to all kinds of airports to reduce the accident rate in the apron and realize the key guarantee for the safety of apron operations.