Autonomous driving technology has always been a very popular research project in scientific theory. Today, with the flourishing development of machine learning and artificial intelligence, we can try to use it in image recognition and vehicle control to achieve more accurate and convenient results. In this thesis, we will use road images to train a convolutional neural network(CNN), for track marking detection, as well as calculations combined with mathematical theory to further carry out road prediction, driving path prediction and vehicle steering angle prediction. In addition to the introduction of mathematical theory and algorithms, we also present some experimental results that verify the theory and algorithms for these predictions using the images of traffics in real roads.