The Design And Implementation Of Lane-keeping System Based On Edge Computing And Deep Learning

Lane keeping is one of the basic functions of autonomous driving,which can prevent traffic accidents caused by lane departure.Autonomous driving is a technology platform on which cross-disciplinary technologies can affect each other and achieve the same goals.More and more breakthroughs have been made in the field of deep learning technology in recent years,which is mainly attributed to the development of hardware computing ability and the large increase of training data samples in the era of big data.Many researchers are trying to apply deep learning autonomous driving to realize the functions of lane detection,obstacle avoidance and lane keeping.The thesis presents the design of a lane-keeping system with the features of data collection,transmitting and model training,which is based on the software architecture of edge computing and the application of the end-to-end convolutional neural network(CNN)to realize the function of lane keeping on a model car.A model car equipped with a monocular camera and controlled by Raspberry Pi(RPi)is designed to realize the lane-keeping system.The Raspberry Pi controlled car can collect and annotate road images automatically by remote control' and then transmit the annotated image data to cloud server via the transmission protocol of Internet of Things(IoT).Convolutional neural network is applied to train the model on the cloud server,thus realizing the lane-keeping function on the model car.The collection and annotation of data sets in the field of autonomous driving is a tedious work,so the thesis proposes an edge computing framework applying to transmitting the training data to save the time and cost for data collecting and annotating.The proposal improved the way of data collecting and annotating on the software architecture and extended the deep learning data sets in autonomous driving.The lane-keeping system is applied on the plantform of Raspberry Pi controlled car.The experimental results show that the system can collect road data and annotate the image automatically.The collected data can be transmitted to the edge server as deep learning data samples.The model car can realize the function of lane-keeping on the road marked by black lines after loading the model that is trained by deep learning.