可控特征点对以及RANSAC的监视系统外参估算方法

为了提高基于机器视觉的监控系统的精度,立体视觉信息作为一种鲁棒性很高的参数被引入监控算法中.立体视觉信息的引入使得外参求解成为影响监控结果的一个重要环节.现存的应用于监视系统的外参自动估算方法均采用立体视觉中外参求解的方法.这些方法普遍存在特征点的数目完全由场景决定特点.为了解决在大部分监视系统应用的场景中,因为背景较为空旷、基线较长同时相机对之间的旋转量较大,由场景提供的特征点数量非常有限,导致计算结果并不理想这一问题,提出了一种基于视频序列的RANSAC外参求解方法.该方法不仅利用RANSAC算法有效的剔除外点,而且根据可控特征点对的动态性,利用时间上的积累获得了充足的位置可控、数量可控的可控特征点对,并据此得到精度较高的外参结果.根据实验,该方法可以将三维重建误差降低近50%.

External parameters calculation for surveillance systems based on controllable feature points and RANSAC algorithm

To solve the problem that the number of obtained feature points is usually not enough to get an ac-curate result for surveillance systems,which is caused by few objects with lots of feature points,the long base-line and the large rotation of cameras,a new algorithm is developed,in which the RANSAC external parame-ters calculation is completed based on videos.This algorithm can eliminate the external points by using RANSAC algorithm,and achieve the external parameters with high accuracy according to the dynamic effect of the controllable feature points.Experimental result shows that the error of reconstruction can be reduced by al-most 50% by this method.