鱼眼图像多角度校正方法研究

Research on multi-angle correction method of images taken by fisheye lens

鱼眼镜头凭借超大视角而被广泛应用于水下机器人,然而目前很多方法对于远离中心部分的图像无法做到良好的校正,边缘图像有不同程度的失真和丢失,并且不符合人眼最佳视觉范围在水平方向20°、垂直方向15°的规律。为此,文中提出一种基于人眼特性的改进球面模型的校正方法。该方法能够在校正边缘区域图像的同时较好地满足人眼观察特性。首先,将二维图像平面映射到三维球坐标系中;然后,对坐标系进行旋转变换,通过旋转矩阵将需要校正的边缘部分移动至视野中央;最后,将三维坐标系中的点逆映射至二维平面。实验结果表明,改进后的方法能够对处于鱼眼镜头各角度方向的图像进行良好的校正,且能够较好地保留图像信息,减少去畸变变换中边缘和形态等信息的丢失。

Fisheye lenses are widely used in underwater robots because of their oversized viewing angle.For many methods,the image edges far away from the image center cannot be well corrected,the image edges have different degrees of distortion and loss,and the rule that the best visual range of human eyes is 20°in the horizontal direction and 15°in the vertical direction cannot be met.Therefore,a correction method of improved spherical model based on human eye characteristics is proposed.In this method,the human eye observation characteristics can be nicely met when correcting the edge areas of the image.The two-dimensional image plane is mapped to the 3D spherical coordinate system,the coordinate system is rotated and transformed,and the edge part to be corrected is moved to the center of the field of view by mans of the rotation matrix,and the points in the 3D coordinate system are reversely mapped to the two-dimensional plane.The experimental results show that the improved method can well correct the Omni-directional parts in the image taken by the fisheye lens,and can well retain the image information well,and reduce the loss of edge and shape information in the process of the distortion transformation.