摄像机位姿的高精度快速求解  

An accurate and fast algorithm for camera pose estimation

作　　者：李书杰[1] 刘晓平[1] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009

出　　处：《中国图象图形学报》2014年第1期20-27,共8页Journal of Image and Graphics

基　　金：安徽省教育厅自然科学研究重大项目(KJ2013ZD02);高等学校博士学科点专项科研基金项目(20120111110003);中央高校基本科研业务费专项资金(2013HGBH0001;2013HGBZ0170)

摘　　要：目的针对对应点个数大于等于6的摄像机位姿估计问题,提出一种既适用于已标定也适用于未标定摄像机的时间复杂度为O(n)的高精度快速算法。方法首先选取4个非共面虚拟控制点,并根据空间点和虚拟控制点的空间关系以及空间点的图像建立线性方程组,以此求解虚拟控制点的图像坐标及摄像机内参,再由POSIT(pose from orthography and scaling with iterations)算法根据虚拟控制点及其图像坐标求解旋转矩阵和平移向量。结果模拟数据实验和真实图像实验表明该算法时间复杂度和计算精度均优于现有的已标定摄像机位姿的高精度快速求解算法EPnP(efficient perspective-n-point)。结论该算法能够同时估计摄像机内外参数,而且比现有算法具有更好的速度和精度。Objective Aiming at the camera pose estimation problem, with the corresponding points number greater than six, we propose an accurate and fast algorithm which has a time complexity of O（n） and is applicable to both： calibrated cameras and uncalibrated cameras. Method First, four non-coplanar virtual control points are selected and linear equations according to the relationship between space points and virtual control points and the image of space points are being setup. Then the image coordinates of the virtual control points and the camera internal parameters are solved by these equations. Second, the rotation matrix and the translation vector are solved by the POSIT （ pose from orthography and scaling with iter- ations） algorithm according to virtual control points and their image points. Result Experimental results with simulative data and real images indicate that the time complexity and calculation precision of the proposed algorithm are better than EPnP （efficient perspective-n-point） algorithm which is the existing accurate and fast pose estimation algorithm for calibrated camera. Conclusion The proposed algorithm can estimate both camera internal and external parameters, and time complexi- ty and calculation precision are better then existing methods.

关 键 词：已标定摄像机 未标定摄像机 PNP问题 EPnP算法 POSIT算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]