天问一号着陆器双目视觉避障技术

对天问一号双目视觉避障技术的组成、方案及原理进行了介绍，通过图像障碍识别和双目立体地形感知完成天问一号降落过程中的安全着陆点自动选取。通过高分辨率双目匹配和亚像素插值处理，提高远距离双目测量精度；在计算资源严格受限的条件下，采用FPGA作为硬件加速平台，通过功能模块的并行化和流水线设计，提高计算视差和三维信息的速度；为克服在轨恶劣环境导致的参数变化，设计在轨自标定方法实现双目对正参数表的自主更新；开展了地面试验和在轨应用，验证了双目视觉避障技术的有效性，实现2 K×2 K高分辨率双目稠密匹配，匹配用时1 s；三维地形测量精度达到6.4 cm@70 m，坡度识别精度0.5°；双目避障系统成功完成了在轨自标定处理，在轨图像能够有效恢复天问一号着陆点区域的三维地形。

Evaluation for Stereo vision Hazard Avoidance Technology of Tianwen 1 Lander

The composition, scheme and principle of the stereo vision hazard avoidance technology for Tianwen 1 are introduced. Automatic selection of safe landing sites during the landing has been finished by image disturbance recognition and stereo terrain perception. High resolution stereo matching and sub pixel interpolation are used to improve the accuracy of long range stereo measurement. And limited by the computing resource, the FPGA is chosen as a hardware acceleration platform, and the speed of calculating parallax as well as 3D information are improved by the parallelization and pipeline design of the function module. To overcome the changes of parameters caused by the harsh environment on orbit, self calibration method is designed to update the stereo correction parameter table. The algorithm verification and on orbit applications have been carried out and the effectiveness of this technology has been verified. The 2K×2K high resolution dense stereo matching is realized by taking 1 s and the accuracy of 3D topographic survey and slope recognition reaches up to 6.7 cm@70 m and 0.5°, respectively. The system has successfully finished the on orbit self calibration, and the on orbit images can effectively restore the 3D terrain of the landing area.