基于多层前馈型神经元网络模型的摄像机定标算法

Camera Calibration Algorithm Based on Multiple-Layer-Forward-Neural Network

针对传统线性、非线性摄像机定标算法的鲁棒性 ,提出了基于多层前馈型神经元 ( MLFN)网络模型来替代传统精确的摄像机定标数学模型 .MLFN网络模型作为层状网络可实现任意维复杂的输入 /输出映射 ,对于无需计算摄像机内、外参数的应用场合 ,该模型提供了一种实用且有较好鲁棒性的摄像机定标算法 ,同时为了补偿摄像机非线性畸变 ,把图像按畸变程度分割成两个区域 ,分别建立各自基于 MLFN网络的摄像机定标模型 .实验表明 ,该方法有效补偿了畸变 ,并提高了模型精度 .给出了基于 MLFN网络模型摄像机定标算法实验 ,验证了该模型的有效性 .

Addressing the robustness of classic linear or nonlinear camera calibration algorithms, this paper put out a novel camera calibration algorithm based on Multiple-Layer-Forward-Neural (MLFN) network. As MLFN network is a multiple layer network, it can map any dimensions of complicated input to output. It presents a useful and well robust camera calibration algorithm when calculating internal and exterior parameters. Furthermore, to compensate the camera's nonlinear distortion, the image is divided into two areas, where MLFN network models will be built on each of them. Experiments indicate that the method can effectively compensate for the distortion of cameras and camera calibration based on the MLFN network proves the effectiveness of this algorithm.