Applying Neural Networks in 3-d Vision

This work presents a proposal for camera calibration using neural networks of the type multilayer perceptron. Camera calibration is employed in 3-D computer vision for pose determination and it requires a solution of non-linear system of equations. By employing neural network, it becomes unnecessary to know the parameters of the cameras, such as focus, distortions and aspects referent to the geometry of the system. Camera simulations and real experiments are used to demonstrate and evaluate the proposed. Key-Words: Computer vision, camera calibration, neural networks, stereo vision, 3-D vision, 3-D image acquisition.