Bird’s-Eye View Vision System for Vehicle Surrounding Monitoring
Liu YC., Lin KY., Chen YS. (2008) Bird’s-Eye View Vision System for Vehicle Surrounding Monitoring. In: Sommer G., Klette R. (eds) Robot Vision. RobVis 2008. Lecture Notes in Computer Science, vol 4931. Springer, Berlin, Heidelberg
本文提出使用6个鱼眼相机 six fisheye cameras 拼接得到环车全景图
系统的流程图如下所示
2.1 Fisheye Camera Calibration 鱼眼相机标定 这里我们使用 Field of View (FOV) 作为 radial distortion 的模型，使用张正友标定方法进行标定，使用 Levenberg-Marquardt method 做整体参数优化
最后进行图像矫正 we rectify the fisheye images into perspective ones by using the FOV model with the estimated parameter values
2.2 Planar Image Alignment with Homography 平面图像的对齐 将所有图像的坐标系统一到一个坐标系下面 all images can be transferred into the same coordinate system and then be registered together
2.3 Optimal Seam Selection 当我们将图像统一到一个 planar surface 上，下一步就是选择哪个图像的像素用于合成 最终的拼接图像 Once having registered all images on a planar surface, the next step is to decide which pixels should contribute to the final composite.
我们希望得到一个 artifacts 尽可能少的 拼接图，we select an optimal seam which lessens mis-registration 这里定义了一个基于 residual error 的损失函数 the cost function according to the residual error for the optimal seam selection
