The Design And Implementation Of The Lane Detection System Based On Deep Learning

In recent years, with the development of economy and science and technology, people's living standards improve a lot. As the urban traffic become more convenient, the number of cars increasing sharply, and the number of traffic accidents caused by the cars is also increasing. In order to improve vehicle safety and reducing traffic accidents, automatic driving vehicle research and development came into being in this situation. In the automatic driving vehicle systems, lane detection is an important part of the sensing module in automatic driving vehicle.Vision-based lane detection solution is one of the lane detection methods which is commonly used. Vision Solution is mainly based on the image processing algorithms to detect the lane carriageway road sign area. Due to the many types of lane markers, and the lane markers are always blocked caused by vehicle congestion, highway lane detection is a very challenging task. There may be situations lane corrosive wear, as well as weather and other factors also can give lane detection task a big challenge.Traditional lane detection algorithms are using special filters to detect lane area, and then combined with Hough transform, RANSAC algorithms to find the lane. These algorithms always need to adjust filter operator manually, find adjustment parameters acccording to the street scene characteristics, but the process is work load and results have poor robustness. Algorithms produce a poor detection result when driving environment changes significantly.Based on conventional lane detection algorithm, combined with deep learning, we propose a method of using deep neural network algorithm instead of the traditional manual adjustment filter operator to segment the lane area in driving scene as a instance level, find a way to access each lane markers area information as a format of pixel, and then use the method of least squares regression parameters to find the formula of each lane lane and finally feedback parameter equation. In this paper, we mainly use convolution neural network, the network structure uses a Convolution and Deconvolution symmetrical design to do semantic segmentation works on the image of traffic lane region.In this paper, we use C ++ and Python language to develop the lane detection system, the whole system is divided into modules like lane markers labeling module, lane marking results filter module, the image data pre-processing module, lane segmentation model training module, the lane detection module and the results display module, in addition, after the completion of this paper, algorithm for different lane segmentation model was tested and compared to give a lane detection solution that suits for automatic driving vehicle server configuration.After testing, we propose a lane detection method based on deep neural network that are more universal, better able to adapt to all kinds of various types of highway traffic scenes, algorithms have achieved a higher lane line detection effect on the highway, whether it is a straight road, curved road, backlit scenes, vehicle blocking more scenes, the segmentation algorithm can detect the lane area and has a strong robustness.