基于BP神经网络的物镜畸变改正模型精度探讨

The Discussion on the Precision of Lens Distortion Model Based on BP Neural Networks

摘要: 非量测数码相机的检校是近景摄影测量研究的热点问题。本文基于标准图像和畸变图像，采用BP神经网络建立物镜畸变改正模型，针对普通数码相机图像的像点坐标进行改正，并与传统Brown多项式畸变模型的结果进行对比分析。实验结果表明，BP神经网络改正模型的精度要优于传统多项式物理模型的精度，能够较好地满足高精度的近景摄影测量需要。

Abstract: Non-metric digital camera calibration has become a research hotspot in the field of close-range photogrammetry. Based on the standard image and distorted image, this paper has corrected the point coordinates of the non-metric digital camera image by using BP neural network to establish the model of objective lens distortion correction, and the point coordinates were compared with the points corrected based on traditional polynomial distortion models proposed by Mr Brown. The result showed that the accuracy of BP neural network model would be higher and could meet the requirement of the close-range photogrammetry of high precision.