Research On Lane Line Detection Based On Deep Learning

With the rapid development of intelligent interconnection,artificial intelligence and new energy technology,a new era of intelligent cars,which once existed in science fiction movies,has arrived quietly.Technologies related to autonomous driving are also getting a lot of attention.As an important part of traffic elements,lane line plays an extremely important role in the field of automatic driving.Therefore,lane line detection has become one of the key basic technologies to realize automatic driving.Due to the complex transformation of the traffic environment in the real world,the traditional lane line detection methods based on OpenCV are vulnerable to the influence of many factors in the real world,with high detection error rate and low accuracy.In order to meet the requirements of higher standard autonomous driving,many complex lane detection methods based on deep learning have emerged in recent years.However,most of the current lane line detection methods are focused on lane detection from a single image,so the information obtained is limited,The detection effect is not ideal when the vehicle is seriously occluded,the ground is Shadowed and the ground signs are seriously missing.Based on this,a lane line detection model based on deep learning is proposed in this thesis.The model in this thesis is based on the improved CNN+RNN network,and finally combined with the Generative Adversarial Network(GAN)to predict lane lines.Lane lines are solid or dashed lines on the road surface,which can be detected in the image through geometric modeling or semantic segmentation.Combined with RNN,better prediction effect can be achieved.Specifically,lane lines that cannot be accurately detected in the current frame can be inferred by combining the underlying information that exists in the previous and subsequent frames of the image.Critical step is through an encoder CNN piece to extract information from each frame and then receive these by the RNN block has successive frames with time series properties of CNN characteristics,characteristics of learning and lane forecast at the same time,the last type antagonism against network model was generated by learning,very good,ruled out the factor of environment to improve the detection accuracy of the algorithm.In order to verify the accuracy of the algorithm,this thesis uses the Tusmiple Dataset to conduct experiments,and uses the Dataset of challenging scenes produced by Qinnzou et al to conduct tests.Finally,it is shown that the precision of this method is better than other algorithms in lane detection in complex environment.