A Nonlinear Convolutional Neural Network Algorithm for Autonomous Vehicle Lane Line Detection

Abstract

The traditional lane line detection algorithm relies on artificial design features, which has poor robustness and cannot cope with the complex urban street background. With the rise of deep learning technology, the algorithm model with convolutional neural network as the mainstream is widely used in the field of computer vision, which provides a new idea for lane line detection. In order to improve the disadvantages of traditional lane line detection methods that are vulnerable to environmental impact and poor robustness, a nonlinear convolution neural network algorithm for driverless lane line detection is proposed. Firstly, the pretreatment method of extracting the region of interest and enhancing the contrast of lane lines is used to reduce the unnecessary image background and enhance the feature details of the image. Existing deep learning-based lane line detection algorithms still have difficulties. First, accumulated wear and tear will cause lane line to fade and fade; roadside trees and buildings can interfere with the performance of lane line detection algorithm. In addition, compared with the pixels of the whole picture, the lane line pixels are too few, and the deep convolution neural network of layer convolution is easy to lead to the loss of details. In addition, when the traffic flow is large, the lane line is easily blocked, which makes it more difficult to detect the lane line. Then the model is built based on the lane line image features extracted by CNN, and the DBSCAN clustering algorithm is used to post-process the lane line segmentation model; Finally, the least square method is used to fit the quadratic curve of the pixel peak points of the lane line, and the fitting results are regressed to the original image. The experimental results show that the accuracy and recall of the lane line detection model verification set are 91.3% and 90.6%, respectively, indicating that the model has a good segmentation effect. It is proved that the lane line detection method based on CNN combined with post-processing can effectively reduce the defects of artificial experience, and has better robustness and accuracy than the traditional lane line detection method.