Research On Lane Detection And Tracking Method Of Autonomous Vehicle Based On Deep Learning

Lane detection and tracking technology is the key technology to realize the current automobile assistant driving and the future automobile unmanned driving.Deep learning technology has developed rapidly in recent years,and has made outstanding achievements in image recognition and positioning,image segmentation,speech recognition and data prediction.In this paper,a lane detection and tracking method based on depth learning is proposed to solve the problem of lane detection and tracking in driverless vehicle environment perception.The main work is as follows:Firstly,the lane detection data set and driving behavior prediction data set for the actual environment and simulation environment are made.Camera calibration of lane acquisition camera is carried out,and the corresponding relationship among image coordinate system,camera coordinate system and world coordinate system is established.Secondly,a lane detection method based on VGG is proposed.To solve the problem of insufficient information fusion of traditional deep learning methods,a context fusion method of feature map is proposed.Combining this method with VGG network,a lane detection method of VGG-FF and VGG-FFD is proposed respectively.The validity of this method is verified by comparing experiments on the CUlane dataset.Thirdly,a driving behavior prediction method based on Res-RNN is proposed.Based on the VGG-FFD lane detection method mentioned above,in order to simulate the mapping relationship between lane and driving behavior,a driving behavior prediction method based on Res-RNN is proposed.By adding residual propagation mechanism into the traditional recurrent neural network(RNN),a better driving behavior prediction is achieved.The results show that the proposed method with residual propagation is better than the traditional RNN method.Fourth,a deep learning-based lane tracking method is proposed.On the basis of the comprehensive analysis of lane line tracking method,the lane line detection method VGGFFD is combined with the driving behavior prediction method Res-RNN,and a lane tracking method VGG-RNN based on deep learning is proposed,which can simulate the function of human driving experience to directly output the correct steering wheel angle and speed based on the road image in front of the car.Finally,this paper builds a simulation platform of driverless vehicle based on Webots.The lane tracking method is simulated and verified by experiments on the real environment data set.The results show that the lane tracking method proposed in this paper can realize the lane tracking of driverless cars.