Orientation Estimation of Non-cooperative Aircraft Based on Residual Network

The end-to-end orientation estimation method based on Convolutional Neural Networks (CNN) is proposed for non-cooperative situations where aircraft can’t send information to each other. In this paper, ResNet50 is used as the backbone network, using Euler angles and quaternions as labels to train the network. The dataset used to train CNN is generated through 3Dmax. The experimental results show that the orientation estimation of non-cooperative aircraft can be achieved through regression method, and it has high orientation estimation accuracy. The orientation estimation accuracy using two labels is compared. The minimum standard deviation of pitch angle, yaw angle and roll angle error is 2.005°, 1.117° and 2.012° respectively, and the minimum mean absolute error is 1.369°, 0.917° and 1.428° respectively. The minimum standard deviation of quaternion rotation angle is 0.375°, and the minimum mean absolute error is 1.292°.