Camera calibration is of great significance in computer vision technology. Large distortion binocular camera is notoriously hard to calibrate accurately using traditional calibration methods. In this paper, we present a new calibration method for large distortion binocular camera based on fully connected neural network. Similarly to planar calibration, it relies on multiple images taken by binocular camera. Compared with traditional calibration methods, the number of images do not increase significantly. Instead of building complex analytical models, we only take some images to train quickly and efficiently the fully connected neural network for different binocular camera systems. The neural network fit accurately the nonlinear mapping relationship of binocular vision imaging model. Experiments show that, compared with traditional calibration methods, this approach calibrate large distortion binocular camera with higher accuracy and stability.