双目摄像机定标的神经网络方法

Camera calibration algorithm based on BP neural network

研究基于反向传播神经网络的摄像机双目立体视觉定标新方法。传统方法基于三角测量原理技术,会带入成像畸变非线性误差,而这种新方法可以消除非线性因素的影响。该方法利用了B P网络良好的非线性映射能力以及学习、泛化能力,通过采用高精度样本数据训练B P网络,最终建立起立体视觉定标的网络模型。由于不需要考虑视觉模型误差、光学调整误差、广角畸变等因素对视觉检测系统测量精度的影响,因而能够有效地克服常规建模方法的不足,保证了检测系统具有较高的精度。

Based on Back Propagation （BP） neural network, a new method of binocular calibration algorithm is presented. The traditional method for stereo vision calibration algorithm based on the principle of trigonometry introduces the nonlinear error because of imaging distortion, and the new method can avoid the influence of nonlinearity. This method uses the ability of nonlinear mapping, studding and generalization of BP ANN（Altificial Neural Network） to establish the mapping relationship between a world coordinate system and an image coordinate system. By ignoring the vision modeling error, and also allowing the exist of optical adjustment error, it overcomes the disadvantages of general methods efficiently, and ensures the vision inspection to have much high measurement accuracy.