基于信用度分类遮挡问题解决方法

Using Credit Coefficients to Solve Occlusion Problems in the Stereo Vision

在基于立体视觉的立式风洞飞机尾旋姿态测量过程中,遮挡问题是困扰测量的一个难题。遮挡问题的常用解决方法需要恢复连续的被遮挡空间,计算量大,且不容易保证精度。针对特征点测量的遮挡问题,本文提出一种基于信用度分类的解决方法,在被遮挡空间搜索感兴趣点的最佳近似,并且通过引入信用度因子的组合计算,控制和提高最佳近似点的计算精度。文中分析了辅助计算点的测量误差所带来的搜索定位困难,信用度因子的引入可以弱化不精确辅助点对搜索结果的影响,使搜索过程主要依赖可靠的点进行。最后,通过仿真实验分析了该方法的可靠性,并说明其在飞机尾旋运动测量实验中对测量结果的有效改善。

In the course of determining the aircraft model attitude in the spin time based on the stereo vision, the occlusion causes much trouble. A new way based on the credit coefficients is shown and it is mainly applicable to the problems with feature matching. In the occlusion problems, it is always the hardest to get the best optimal estimation of the occluded parts. The credit coefficients can weaken the effect of the features with poor definition. So the course for searching the occluded features is mainly based on the reliable features. Simulations and experiments on real images validate the reliability of our method and the results are good enough for the aircraft spin experiment.