基于优化YOLO方法机场跑道目标检测  

Runway Target Detection Based on Optimized YOLO Method

作　　者：蔡成涛[1] 吴科君 严勇杰 CAI Chengtao;WU Kejun;YAN YongJie(College o{ Automation,Harbin Engineering University,Harbin 150001,China;State Key Laboratory of Air Traffic Management System and Technology,Nanjing 210007,China)

机构地区：[1]哈尔滨工程大学自动化学院,哈尔滨150001 [2]空中交通管理系统与技术国家重点实验室,南京210007

出　　处：《指挥信息系统与技术》2018年第3期37-41,共5页Command Information System and Technology

基　　金：空中交通管理系统与技术国家重点实验室开放基金(SKLATM201708)资助项目

摘　　要：为提供机场调度决策依据,需对机场跑道上的飞机和车辆等目标进行准确和快速检测。提出了基于优化YOLO框架的机场目标检测方法。根据监控系统拍摄角度得到图像投影,调整YOLO框架的检测网络结构,并在网络中增加卷积层,使检测器适应监控系统的投影成像,从而改善检测效果。构造机场模拟环境进行仿真试验。试验结果表明,在预设30°摄像头俯拍角度下,基于优化YOLO方法的机场检测目标帧率达26帧/s,准确率达91%以上,且对环境具有一定鲁棒性,可满足机场跑道入侵检测实时性要求,为机场塔台调度人员提供辅助决策依据。To provide the basis for airport dispatching decision-making,the targets on airport,such as aircrafts and vehicles on the runway,need to be detected accurately and rapidly.An airport target detection method based on optimized YOLO framework is proposed.The image projections are obtained according to placement angle of the video monitoring system,and the detection network based on the YOLO framework is modified.By adding the convolution layer to the network,the detector is adapted to the projection imaging of the monitoring system,thus obtaining a better detection effect.The airport simulation environment is constructed for the simulation experiment,the experimental result shows that at the preset angle of 30°,the detection method based on the optimized YOLO can achieve over 91% of the detection accuracy,when the target frame rate is 26 FPS.It has some robustness to the environment.Thus,it can provide a reliable basis for decision-making to the airport tower scheduling personnel.

关 键 词：机场安全 目标检测 深度学习 YOLO方法 跑道入侵 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]