Monocular Visual Pose Measurement for Autonomous Landing in Unknown Environments

An airborne monocular visual pose measurement method is proposed for autonomous landing in unknown environment, as there is currently no effective technical method to perform such tasks. The method begins by designing a multi-task neural network to extract semantic segmentation, depth and slope information from monocular images. Next, a 2D grid-scene information map is then constructed to evaluate candidate regions. Candidate regions are identified from this map, and Radon transform operation is applied to search for suitable landing regions. The optimal landing region is then chosen based on the characteristics of the aircraft and represented using keypoints. This representation facilitates precise relative pose measurement between the aircraft and the optimal landing region. To validate the method, the absolute scale of the environment is obtained by consecutive monocular images and tested on both synthetic and real images. The results confirm that this method ensures efficient and accurate autonomous landing measurements.