摘要
摄像机标定是机器视觉中最重要的环节之一,传统标定方法运算量大、计算复杂,非常繁琐。为解决标定存在的若干问题,提出基于改进神经网络的双目视觉摄像机标定方法。通过对双目摄像机有效模型分析,建立空间点图像坐标与世界坐标非线性映射关系,同时引入自适应学习算法,实现隐层神经元的自适应选取,并且在创建网络模型前对样本数据进行归一化处理,提前终止策略,使网络泛化能力得到极大改善。通过与经典标定方法进行比较,表明基于改进型神经网络标定方法能获得较好的双目标定精度。
Computer vision has being applied widely at industrial, military and transportation. Camera calibration is one of the most important aspects of computer vision, but traditional calibration methods are comparatively complicated, and the arithmetic amounts are big. To solve some issues in calibration, this paper proposed the camera calibration method based on binocular vision improved neural network. In this method, nonlinear mapping between image coordinates and world coordinates is set up through analysis of binocular camera model ; then introducing an adaptive learning algorithm, the adaptive selection of hidden layer neuron is realized; before creating the network model, the data sample is normalized, therefore, recognition ability of network is improved. Compared with traditional calibration methods, experimental results show that the proposed binocular calibration method based on improved neural network could obtain high accuracy.
出处
《西南科技大学学报》
CAS
2013年第4期66-70,共5页
Journal of Southwest University of Science and Technology
基金
国防基础科研计划资助项目(B3120110005)
关键词
摄像机标定
双目视觉
神经网络
自适应性
Camera calibration
Binocular vision
Neural network
Adaptation
