关键词: 相机标定, BP神经网络, 遗传算法, 角点检测, 同名角点匹配

Abstract: The BP neural network has the problems of low accuracy and poor convergence in the binocular camera calibration processes. A method was proposed to solve these problems to complete the calibration of binocular camera based on BP neural network optimized by improved genetic algorithm. First, the Trajkovic operator with multi-lattice fusion algorithm was proposed for corner detection. A matching algorithm of homonymous corner combining point-to-point spatial mapping and grid motion statistics was proposed. Then, the pixel value of the homonymous corner was extracted and the actual 3D coordinate value was calculated, and the crossover and mutation probability and selection operator of genetic algorithm were improved. Finally, the improved genetic algorithm was used to optimize the BP neural network, and the pixel value and the 3D coordinate value were used as the input and output of the BP neural network respectively, so as to complete the calibration of binocular camera. The results show that average calibration prediction errors before and after optimization are as 0.66 mm and 0.08 mm respectively, and the average calibration prediction error was reduced by 88%. The number of calibration test iterations before and after optimization is as 736 and 169 respectively, and the iteration speed is increased by 3.4 times after optimization. The improved genetic algorithm optimizes the BP network and achieves good results, which basically meets the requirements of binocular camera calibration. 

Key words: camera calibration, BP neural network, genetic algorithm, corner detection, homonymous corner matching