基于未标定普通相机的全景图像拼接方法

Panoramic Stitching Method Based on Uncalibrated Ordinary Camera

提出了一种利用普通相机拍摄的图片来构造全景图像的方法。首先利用ORB特征匹配算法对相邻图像进行匹配,并利用匹配点计算图像间的单应性变换矩阵;其次,对单应性矩阵进行分解,估计出摄像机内参数矩阵及图像间的旋转变换矩阵,并为保持全局一致性,建立全局旋转矩阵以消除拍摄过程中产生的抖动和倾斜;最后,建立柱面环境图,将拼接图像卷绕至柱面,并对拼接后的图像进行融合处理生成360°柱面全景图。本算法使用未标定的普通相机拍摄的图片即可生成柱面全景图像。实验结果表明该方法的拼接效果好,有较强的实用价值。

A panorama constructing method based on images taken by ordinary cameras was proposed. Firstly, image features were matched using the ORB algorithm and match-pairs were used to compute homographies between images. Secondly, homographies were decomposed to calibration matrices and rotation matrices. In order to keep global consistency, the global rotation matrices was constructed to eliminate the effect of panning and tilting camera. Lastly, cylindrical environmental map was built. The stitching images were warped to environmental map and blended to generate the final panorama. The proposed method used images taken by ordinary cameras to generate cylindrical panorama. The experiment shows that the proposed algorithm has a good effect of stitching and practicality.