When using visual and other optical means to navigate during the asteroid exploration flight

the aspherical surface illumination changes will affect the stability of the visual image features

so as to affect the estimation of the relative pose. A pose estimation method based on homography matrix is proposed aiming at this problem. The method realizes estimation of the homography matrix with a deep learning framework based on a recurrent convolutional network-convolutional neural network（ RNN-CNN）. The experimental results show that the network improves the precision of the homography matrix estimation under illumination variation. The method is applied to estimate poses of the asteroid surface images sequence

which proves the effectiveness of the method and its superiority in precision and efficiency over traditional methods.