摘要
针对摄像机线性标定准确程度不高与非线性标定计算复杂等问题,以己知的三维信息为基础推导二维像素点对信息,提出了基于神经网络的三目立体视觉摄像机标定方法,构建了改进的BP神经网络模型,对比分析了两种摄像机标定方法的像素点对均方根误差。结果表明:采用改进的BP神经网络能够避免对摄像机进行非线性建模,有利于提高标定精度,增加系统的灵活性,更具有实际意义。
To overcome the shortcoming of the linear and non linear calibration in the processing of camera calibration,the 2 D pixels information was deduced based on the foregone 3 D information, a new intelligent calibration algorithm was presented based on neural network and an improved BP network model was constructed, and pixels point errors in two sort calibration methods were compared. The experiment results indicate that the improved BP network model can avoid non-linear modeling and enhance the calibration precision and the flexibility, which has real significance.
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2007年第2期391-395,共5页
Journal of Jilin University:Engineering and Technology Edition
基金
国家自然科学基金资助项目(50475011)
教育部优秀青年教师资助计划项目
关键词
自动控制技术
工程机器人
立体视觉
图像处理
神经网络
摄像机标定
automatic control technology
construction robot
stereo vision
image processing
neuralnetwork
camera calibration
