Line Detecting System Based On Hough Tansform And Convolutional Neural Network

In recent years,intelligent vehicle has attracted more and more attentions from various researchers,among those researches,detecting line based on vision has aroused most interest.Detecting line and tracking line are important for the sake of autonomous driving,regardless of adopting infrared sensor,or using machine learning algorithms,the purpose of detecting line is to provide solid information for controlling the smart car,in other words,the driving environment should be well perceived by the agent.With the background of line-detecting and tracking for autonomous driving,this paper proposes a line-detecting and control system which will fulfill the purpose of improve the stability and safety of autonomous driving.The system proposed in this paper could be divided into several modules: image preprocess module,image classification module and car controlling module.Susceptible to environment change,road image is easily to be affected by noises.For a better accuracy of line-detecting,this paper using Canny edge detect and Hough transform to ease the noises,making the line more stand out and filtering out other noncorrelated information.Then a convolutional neural network with 6 layers was applied to classify those images,the accuracy has reached to 90.08%.In the last,Q-learning was adopted for controlling system,making the agent has the ability to learn from different environment in an interactive way.In the last part of this thesis,the lane-detecting system was verified.All composing modules were tested individually,all of them met the requirements and passed all the tests.Comparison was made between the system proposed in this thesis and other lane-detecting system.