变异机制粒子群优化的摄像机内参数校准  

Mutation mechanism particle swarm optimization for camera internal parameter calibration

作　　者：周婧[1] 张小宝 白云龙 ZHOU Jing;ZHANG Xiao-bao;BAI Yun-long(College of Information Technology,Jilin Agricultural University,Changchun 130118,China)

机构地区：[1]吉林农业大学信息技术学院

出　　处：《光学精密工程》2019年第8期1745-1753,共9页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.31801753);吉林省教育厅科学技术研究项目(No.2015202)

摘　　要：针对大空间单目视觉系统中摄像机内参数校准精度对整体测量精度影响较大这一问题,本文提出一种基于变异机制粒子群优化(MMPSO)算法的摄像机内参数虚拟三维校准方法。该方法基于分阶段最优化思路,通过建立摄像机成像模型对摄像机外参数及部分内参数进行初始值估计,再通过MMPSO算法对内参数进行优化校准确定最终的结果。实验中为了提供精确的校准控制点,搭建了校准硬件平台,将红外发光二极管固定于三坐标测量机测头上并跟随测头移动,构造一个大空间虚拟三维校准板。实验结果表明:主要的10个内参数均达到测量精度要求的数量级,验证了该方法的有效性。通过单目视觉坐标测量系统对两种校准方法所得结果进行等距测量实验,基于Janne Heikkila的三维校准法的总体标准差为0.112 mm,基于MMPSO算法的虚拟三维校准法的总体标准差为0.084 mm。通过对比实测数据标准差,可以证明本文提出的校准方法稳定性更好,精度更高。该方法能够满足大空间单目视觉坐标测量系统对摄像机内参数精度的要求,对视觉坐标测量技术领域中的摄像机校准等非线性优化问题具有一定指导作用。To solve the problem that the calibration accuracy of internal parameters of the camera in the large-size single-camera vision system has great influence on the overall measurement accuracy,this paper presents a virtual stereo calibration method for the internal parameters of the camera based on the Mutation Mechanism Particle Swarm Optimization(MMPSO)algorithm.The method is based on a two-stage optimal strategy.Firstly,a camera imaging model is established to estimate the initial values of the external parameters and some internal parameters.Then,the internal parameters are optimized and calibrated by the MMPSO algorithm to determine the final result.To provide accurate calibration control points,a calibration hardware platform was built.An infrared light-emitting diode was fixed on the measuring head of a three-coordinate measuring machine(CMM),which drove the diode to move,and a large-space virtual three-dimensional calibration board was constructed.The experimental results showed that all of the 10 main internal parameters reached the order of magnitude requested by the measurement accuracy,which validated the effectiveness of the method.The results of two calibration methods were measured by equidistant measurement with the single-camera vision coordinate measurement system.The population standard deviation of the three-dimensional calibration method of Janne Heikkila was 0.112 mm,but the population standard deviation of the virtual stereo calibration method based on the MMPSO algorithm was 0.084 mm.The comparison of the standard deviations of the measured data proves that the proposed calibration method is more stable and accurate.This method can meet the requirements of the large-space single-camera vision measurement system for the accuracy of camera parameters,and it has a certain guiding effect on nonlinear optimization problems such as camera calibration in the field of visual coordinate measurement technology.

关 键 词：视觉测量 摄像机模型 变异机制粒子群优化算法 内参数校准 

分 类 号：TH741[机械工程—光学工程]