着陆避障中双目立体匹配方法与仿真

摘要： 在着陆避障过程中，利用双目相机拍得的图片进行立体匹配得到稠密的地形高程图是一种有效的方式.文中对绝对误差之和(sum of absolute differences, SAD)、全Census变换以及稀疏Census变换立体匹配方法原理进行说明，针对SAD算法无法处理左右图像亮度不一致的情况提出了对匹配窗口进行灰度规范化以及加入LOG滤波的改进措施，并且对上述方法分别应用数据库图像和沙盘拍图进行仿真测试及比较，从匹配错误率和实时性的角度分析各算法在地外星体着陆这种特殊场景中的使用性能.

Matching Techniques and Simulation for Hazard Detection During Planetary Landing

Abstract： During planetary landing, it is an effective method to achieve hazard detection and avoidance by analyzing dense digital elevation map getting from two pictures taken by binocular cameras via matching algorithms. The principles of SAD, full census transformation and sparse census stereo matching techniques are explained. And an improved SAD algorithm and LOG filter are proposed to overcome the weakness that a simple SAD cannot handle the condition when two pictures have different illumination. Then the algorithms are simulated and compared via stereo images from Middlebury College and sand table pictures. The matching error rate, the real time and the usability of the different algorithms are discussed in the scene of planetary landing.