Detekce směru pohledu řidiče pomocí konvolučních neuronových sítí

Abstract

This study was focus on driver gaze estimation by using a convolutional neural network. Two approaches were tested. The one which using the external face detector and the second one which not using any external detectors. Also, the various learning parameters of the convolutional neural network were tested like the setting of random cropper, windows size, etc. Two simple datasets were created for learning and testing the convolutional neural network. The first one was created for the learning of the convolutional neural network for head pose detection. The second one was created for the learning of the convolutional neural network for eye gaze. Both of them were created in the indoor environment and have nine classes.