一种灵活的柱面反投影算法

Flexible cylindrical back-projection algorithm

摘要:

传统柱面图像反投影算法依赖于圆柱模型以及标定的精度，且物距发生变化时模型将会失效，需要重新进行标定，过程繁琐，场景适应性不足.为此，提出一种灵活的柱面图像反投影算法.根据透视投影原理，像素当量随距离远近呈二次曲线分布，因此可以通过拟合二次曲线表征其成像特性，对竖直方向和水平方向分别进行处理，进而实现柱面图像反投影变换.当目标物体相对相机发生前后方向和左右方向位移时，通过修正二次曲线的系数适应其成像特性的改变，无需再次标定.实验结果表明，该算法精度与传统方法相当，且在左右移动、一定缩放范围内保持精度不变.本文算法仅需一次标定即可满足目标物体小范围内移动时的图像变换需求，提高了场景适应性.

Abstract:

Traditional cylindrical back-projection algorithm depends on the cylindrical model and calibration accuracy. While object distance changes, a second calibration needed. A new flexible algorithm is proposed to simplify the process and improve the scene adaptability. According to perspective projection theory, pixel equivalent is distributed by conics. The algorithm gets cylindrical imaging characteristics to realize cylindrical back-projection by means of fitting conics in the horizontal and vertical directions separately. It can adapt to the changes of imaging characteristics through amendment of conics coefficient without calibrate again, when the target object moves around relative to the camera. Experiment results show that the proposed algorithm accuracy is about normal for traditional way and remains unchanged while the target object moves around within the scope of a certain scale. The proposed algorithm meets image transformation requirements with calibration only once when the target object moves around on a small scale and scene adaptability is improved.