基于改进YOLOv5的飞机舱门识别与定位方法研究

机场特种车辆的自动靠机是未来智慧机场发展的必然要求，实现自动靠机的关键是对飞机舱门进行准确识别与定位；针对于此问题，提出一种基于改进YOLOv5和单目视觉的舱门识别与定位方法，通过在模型中加入了一种轻量化的卷积注意力模块（CBAM，convolutional block attention module）,提高了算法对飞机舱门的特征提取能力；针对YOLOv5的重复特征提取问题，引入了空间金字塔池化结构（SPPCSPC，spatial pyramid pooling cross stage paritial connection），并改进分组卷积组数为4，提高了算法的检测精度；通过获取候选框中角点的像素，利用空间几何关系，实现了对舱门准确的三维定位。实验结果表明，改进后的YOLOv5算法mAP达到96.5%，相比原有算法提升了5.6%。在舱门前方19 m和1 m处时，实时最大定位误差分别为0.15 m和0.01 m，能够满足特种车辆靠机完成后与舱门保持5-10 cm的安全距离要求。

Research on Aircraft door identification and Locationmethod based on improved YOLOv5

The automatic docking of airport special vehicles is an inevitable requirement for the development of smart airports in the future; The key to achieving automatic docking is to accurately identify and position the aircraft door.Aiming at this problem,proposes a door recognition and position method based on improved YOLOv5 and monocular vision. By adding a lightweight convolutional block attention module (CBAM) to the model, the algorithm improves its ability to extract features from aircraft doors; To solve the problem of repetitive feature extraction in YOLOv5, a spatial pyramid pooling cross stage partial connection (SPPCSPC) is introduced, and the number of group convolution groups is improved to 4, improving the detection accuracy of the algorithm; By obtaining the pixels of corner points in the candidate frame and utilizing spatial geometric relationships, accurate three-dimensional positioning of the aircraft door is achieved. The experimental results show that the improved YOLOv5 algorithm mAP reaches 96.5%, which is 5.6% higher than the original algorithm. At 19 m and 1 m in front of the aircraft door, the real-time maximum positioning error is 0.15 m and 0.01 m, respectively, which can meet the requirements of maintaining a safe distance of 5-10 cm from the aircraft door after the completion of docking of special vehicles.