论文部分内容阅读
提出一种基于神经网络的系统标定方法.通过射影变换及误差补偿方法,建立摄像机图像平面与投影仪图像平面的映射关系,利用该映射关系和标定点的摄像机图像坐标,计算得到相应的投影仪图像坐标;建立三层结构的神经网络,该网络以两个图像坐标为输入,对应的世界坐标为输出,训练样本由得到的标定点的两个图像坐标及其世界坐标组成,采用BP算法训练该网络;训练过程即为神经网络逼近系统模型的过程,训练完成时,系统完成标定.实验表明,与传统的结构光标定方法对比,本文提出的方法简化了建模复杂度和标定过程,提高了标定精度,并具有普遍适应性.
A method of system calibration based on neural network is proposed.The mapping relationship between camera image plane and projector image plane is established by projective transformation and error compensation method.The corresponding projector is calculated by using the mapping relationship and camera image coordinates of calibration point Image coordinates; establish a three-layer structure of the neural network, the network to two image coordinates as input, the corresponding world coordinates for the output of the training sample from the calibration point of the two image coordinates and world coordinates, using BP algorithm training The training process is the process of neural network approximation system model, and the system completes the calibration when the training is completed.The experiment shows that compared with the traditional structured light calibration method, the proposed method simplifies the modeling complexity and calibration process, The calibration accuracy, and has universal adaptability.