采用改进遗传算法优化神经网络的双目相机标定  

Binocular Camera Calibration Based on BP Neural Network Optimized byImproved Genetic Algorithm

作　　者：张峰峰[1,2] 张欣 陈龙 孙立宁 詹蔚[3] ZHANG Fengfeng;ZHANG Xin;CHEN Long;SUN Lining;ZHAN Wei(School of Mechanical and Electrical Engineering,Soochow University,Suzhou,Jiangsu,215006;Collaborative Innovation Center of Suzhou Nano Science and Technology,Soochow University,Suzhou,Jiangsu,215123;Department of Radiation Oncology,The First Affiliated Hospital of Soochow University,Suzhou,Jiangsu,215000)

机构地区：[1]苏州大学机电工程学院,苏州215006 [2]苏州大学苏州纳米科技协同创新中心,苏州215123 [3]苏州大学附属第一医院放疗科,苏州215000

出　　处：《中国机械工程》2021年第12期1423-1431,共9页China Mechanical Engineering

基　　金：国家重点研发计划(2018YFB1307700)。

摘　　要：针对传统BP神经网络在双目相机标定过程中存在的迭代时间长、精度低等问题,提出了基于改进遗传算法优化BP神经网络的方法来完成双目相机标定。使用融合多格算法的Trajkovic算子进行角点检测,利用点对点空间映射和网格运动统计相结合的方法完成同名角点匹配,在此基础上,提取同名角点的像素值并计算其实际的三维坐标值。对遗传算法的交叉和变异概率及选择算子进行改进,利用改进后的遗传算法对BP神经网络进行优化,将像素值和三维坐标值分别作为BP神经网络的输入和输出,进而完成双目相机的标定。实验结果表明:优化前后的平均标定预测精度分别为0.66 mm和0.08 mm,其平均标定预测精度提高了88%。优化前后的标定测试迭代次数分别为736和169,优化后迭代速度提高了3.4倍。改进遗传算法优化BP网络在双目相机标定过程中取得较好的效果,满足了双目相机标定的要求。The BP neural network has the problems of low accuracy and poor convergence in the binocular camera calibration processes.A method was proposed to solve these problems to complete the calibration of binocular camera based on BP neural network optimized by improved genetic algorithm.First,the Trajkovic operator with multi-lattice fusion algorithm was proposed for corner detection.A matching algorithm of homonymous corner combining point-to-point spatial mapping and grid motion statistics was proposed.Then,the pixel value of the homonymous corner was extracted and the actual 3D coordinate value was calculated,and the crossover and mutation probability and selection operator of genetic algorithm were improved.Finally,the improved genetic algorithm was used to optimize the BP neural network,and the pixel value and the 3D coordinate value were used as the input and output of the BP neural network respectively,so as to complete the calibration of binocular camera.The results show that average calibration prediction errors before and after optimization are as 0.66 mm and 0.08 mm respectively,and the average calibration prediction error was reduced by 88%.The number of calibration test iterations before and after optimization is as 736 and 169 respectively,and the iteration speed is increased by 3.4 times after optimization.The improved genetic algorithm optimizes the BP network and achieves good results,which basically meets the requirements of binocular camera calibration.

关 键 词：相机标定 BP神经网络 遗传算法 角点检测 同名角点匹配 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]