In response to the task requirements of high-precision real-time autonomous navigation for the asteroid probe during the descent and landing phase, considering the limited onboard computing resources and the complex terrain environment of the asteroid, this paper proposes a feature anchor point tracking-based autonomous landing navigation method. Firstly, the 3DSIFT algorithm is used to extract initial feature points from the three-dimensional point cloud data of the asteroid, and the spatial distribution feature selection mechanism is adopted to select feature anchors with high recognition degree, providing an accurate three-dimensional reference benchmark for subsequent visual tracking. On this basis, a lightweight convolutional neural network is used to directly detect and track the two-dimensional projections of the feature anchors in the image sequence, avoiding the computationally intensive feature matching steps in traditional methods, while ensuring navigation accuracy and significantly reducing computational complexity. Finally, based on the established 2D-3D point correspondence relationship, the asteroid lander’s pose parameters are accurately estimated using the EPnP method. Through software simulation and semi-physical experiments, it is verified that this method can effectively reduce the allowable computing power of the navigation system while ensuring the accuracy of pose estimation, providing an efficient and reliable solution for the autonomous navigation of the probe during the descent and landing phase of asteroid exploration.
WU et al. (Sun,) studied this question.