自然场景中近景摄影测量相机的自标定方法

2018-12-20

作者：王涛1，董明利1，孙鹏2，王君1，燕必希1 单位：1北京信息科技大学；2机电系统测控北京市重点实验室

摘要：提出了一种高精度的相机自标定方法，用于可见光下大型自然景物的摄影测量和重建。首先，用测量相机拍摄景物的交会图像并通过SURF (speeded up robust features)算法获取图像特征点；其次，使用五点法对各图像进行相对定位定向，并通过随机抽样一致获得最佳定向结果；再次，结合特征描述的相似性和极线约束匹配特征点；最后，使用自标定光束平差技术对相机的内方位参数、图像外方位参数和空间特征点坐标进行最优化。实验结果表明，本方法可以取得同实验室环境下非常接近的相机标定结果。本方法不需要布置人工标志点，利用场景的自然特征即可实现相机的高精度标定，同时可以获得高精度的场景重建结果，适用于大尺寸自然景物的三维高精度重建。

关键词：特征点识别；特征点匹配；相机标定；摄影测量

中图分类号：TG87; TH741文献标志码：A

Selfcalibration Method of Close Range Photogrammetric Camera in Natural Scenes

Wang Tao，Dong Mingli，Sun Peng，Wang Jun，Yan Bixi

Abstract:In this paper，a highprecision camera selfcalibration method is proposed for photogrammetry reconstruction of large natural scenes under visible day light.Firstly，convergent images of the scene are captured and large amount of image feature points are obtained by SURF algorithm；Secondly，the five point method is used for relative orientation and orientation of each image，and the best orientation results are obtained by Random Sample Consensus；Thirdly，the feature points are matched under the constraints of both the similarity of feature description and the epipolar geometry; Finally，the interior orientation parameters of the camera，the parameters of external orientation and the coordinates of the spatial feature points are optimized simultaneously in the selfcalibration bundle adjustment.The experimental results show that this method achieves similar calibration results as in the laboratory environment.This method does not need artificial target points，high precision calibration can be realized using natural features of the scene camera，at the same time can obtain the scene reconstruction results with high accuracy，high precision of 3D reconstruction for large scenes of natural scenery.This method uses natural feature points rather than artificial targets to achieve high accuracy camera calibration.Moreover，this method obtains high accuracy 3D reconstruction of the feature points which makes it suitable forreconstruction and modelling of large scale natural scenes.

Keywords:feature points detection；feature point matching；camera calibration；photogrammetry