基于飞行时间相机的非合作慢旋卫星稠密重建研究

针对非合作慢旋卫星的模型重建问题，提出基于飞行时间(time-of-flight, TOF)相机和同时定位与制图(simultaneous localization and mapping, SLAM)的稠密重建方法。基于预先检测与自适应阈值方法提高旋转提取与描述(oriented fast and rotated brief, ORB)的特征尺度适应性。利用运动度量方法选取关键帧。利用子模型拼接方法加快重建效率。利用仿真环境制作非合作慢旋卫星的数据集。仿真实验结果表明:该方法能够实现长时间稳定地工作，可在3 min内重建出卫星模型的稠密点云，点云密度大于5 000，重建误差小于5 cm。利用机械臂、卫星模型及光学暗室搭建半物理实验系统，表明算法的精度及抗噪声能力基本满足非合作目标感知的任务的需求。

Dense reconstruction of slow-spinning non-cooperative satellite based on time-of-flight camera

To reconstruct a non-cooperative slow-spinning satellite, a time-of-flight (TOF) camera and simultaneous localization and mapping (SLAM) based dense reconstruction method is proposed. First, the feature scale adaptability of oriented fast and rotated brief (ORB) is improved based on a predetection method and an adaptive threshold method. Second, the key frame selection is improved based on the motion measure method. Then, the sub-model registration is used to accelerate the efficiency of reconstruction. The dataset of non-cooperative slow-spinning satellite is made by using simulation environment. The simulation results show that the proposed method works stably for a long time. The density of the reconstructed model is greater than 5 000, the reconstruction error is less than 5 cm, and the reconstruction time is less than 3 min. The semi-physical experimental system is built by using robotic arms, a satellite model and an optical darkroom. The results show that the accuracy and anti-noise ability of the algorithm basically meet the requirements of the perception of non-cooperative targets.