改进遗传模拟退火神经网络的双相机标定算法

Improved dual-camera calibration algorithm based on genetic simulated annealing neural network

针对双目相机标定算法中BP神经网络精度低、收敛性差的问题,提出了一种基于改进遗传模拟退火算法的BP神经网络优化方法(improved genetic simulated annealing algorithm-BP,IGSAA-BP)。该方法通过改进遗传模拟退火算法的适应度扩展,交叉、变异概率和退火准则来改善BP神经网络的性能,为BP神经网络提供了最优权值和阈值。将IGSAA-BP神经网络与BP神经网络和TGSAA-BP神经网络两种相机标定算法进行对比,利用其标定的真实值与预测值相差结果得出结论,3种标定方法进行相机标定的平均标定精度分别为0.02、0.71和0.28 mm。结果表明,IGSAA-BP神经网络可以提高双目相机的标定精度,提高全局寻优能力,加快收敛速度。

Aiming at the problems of low precision and poor convergence of BP neural networks in calibration algorithms of binocular cameras,a new optimization method for BP neural networks based on improved genetic simulated annealing algorithms has been put forward,including IGSAA-BP.This method improves the performance of BP neural network by improving the fitness expansion,crossover,mutation probability and annealing criterion of genetic simulated annealing algorithm,and provides the optimal weight and threshold for BP neural network.By comparing IGSAA-BP neural network with BP neural network and TGSAA-BP neural network,the difference between the real and predicted values of IGSAA-BP neural network and TGSAa-BP neural network,it is concluded that the average calibration accuracy of the three calibration methods for camera calibration is 0.02,0.71 and 0.28 mm,respectively.The results show that IGSAA-BP neural network can improve the calibration accuracy of binocular camera,improve the global optimization ability and accelerate the convergence speed.