Calibration of stereo vision systems by neural networks

In this paper will be analysed a calibration procedure for stereo vision based on neural networks. The neural net is trained using as input pattern the position from two or more CCD cameras of the image points corresponding to the calibration points and as target pattern the actual 3D co-ordinates of the calibration points. The goal is the approximation of the non-linear map which relates the external co-ordinates in 3D space to the CCD locations.