AUTOMATIC VISUAL PERCEPTION USING AN ARTIFICIAL NEURAL NETWORK

摘要

According to a computer-implemented method for training an artificial neural network (8, 8') for automatic visual perception, a training image dataset is provided as an input to the neural network (8, 8') by a training computing unit and the training computing unit is used to receive intrinsic camera calibration data for a non-rectilinear camera (6), initialize a set of trainable mapping parameters depending on the calibration data, map the training image dataset to a two-dimensional manifold (7) according to the set of trainable mapping parameters, apply a convolutional kernel (9) to the mapped training image dataset in order to generate a set of features, generate output data of the neural network (8, 8') depending on the set of features and to modify the set of trainable mapping parameters depending on the output data to train the neural network (8, 8').