Bae Hyunsoo, Kang Wonjin, Lee SukGyu, Kim Youngwoo
Electrical Engineering, Yeungnam University, Gyeungsan 38541, South Korea.
Robotics Engineering, Daegu Gyeongbuk Institute of Science & Technology, Daegu 43008, South Korea.
Rev Sci Instrum. 2016 Dec;87(12):123109. doi: 10.1063/1.4972108.
This paper proposes a new piecewise linear omnidirectional image registration method. The proposed method segments an image captured by multiple cameras into 2D segments defined by feature points of the image and then stitches each segment geometrically by considering the inclination of the segment in the 3D space. Depending on the intended use of image registration, the proposed method can be used to improve image registration accuracy or reduce the computation time in image registration because the trade-off between the computation time and image registration accuracy can be controlled for. In general, nonlinear image registration methods have been used in 3D omnidirectional image registration processes to reduce image distortion by camera lenses. The proposed method depends on a linear transformation process for omnidirectional image registration, and therefore it can enhance the effectiveness of the geometry recognition process, increase image registration accuracy by increasing the number of cameras or feature points of each image, increase the image registration speed by reducing the number of cameras or feature points of each image, and provide simultaneous information on shapes and colors of captured objects.
本文提出了一种新的分段线性全向图像配准方法。该方法将由多个相机捕获的图像分割为由图像特征点定义的二维段，然后通过考虑段在三维空间中的倾斜度对每个段进行几何拼接。根据图像配准的预期用途，该方法可用于提高图像配准精度或减少图像配准中的计算时间，因为可以控制计算时间和图像配准精度之间的权衡。一般来说，非线性图像配准方法已用于三维全向图像配准过程中，以减少相机镜头造成的图像失真。所提出的方法依赖于全向图像配准的线性变换过程，因此它可以提高几何识别过程的有效性，通过增加相机数量或每个图像的特征点数量来提高图像配准精度，通过减少相机数量或每个图像的特征点数量来提高图像配准速度，并提供关于捕获对象的形状和颜色的同步信息。
