Camera calibration correction method based on multiple neural networks

• Main Authors: WU DANLING, TANG SHIXIAN
• Format: Patent
• Language: Chinese; English
• Published: 07.03.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

According to the method, based on linear parameters, a BP neural network is used, nonlinear parameters use a neural network optimized by an ant colony algorithm with relatively high nonlinear fitting capability to obtain internal and external parameters of a camera, outputs of the BP and GA-BP neural networks are used as inputs of a convolutional neural network AlexNet, the AlexNet adopts a dropout strategy and an activation function (ReLU) function to relieve an overfitting problem, and the algorithm is applied to a camera. According to the method, camera calibration and distortion image correction can be autonomously learned in various scenes and complex environments, the operation simplicity and applicability of the camera calibration method are improved, the accuracy and reliability of camera calibration are improved, and the camera calibration requirement is met. 本发明公开了一种基于线性参数使用BP神经网络、非线性参数使用非线性拟合能力较强的蚁群算法优化后的神经网络获取相机的内外参数，将BP、GA-BP神经网络的输出，作为卷积神经网络AlexNet的输入，AlexNet采用dropout策略和激活函数(ReLU)函数缓解过拟合问题，能够在多种场