Research On Detection And Recognition Method Of Multi Type Lane Lines

Lane detection and various road traffic marking recognition,as the key problems in the field of intelligent driving,are widely used in intelligent traffic systems,advanced driver assistance systems and lane departure warning systems.In the face of the increasing number of vehicles,high traffic accidents and the increasingly complex road traffic environment,it is of great significance to carry out the research on lane detection and road traffic marking recognition methods with real-time application as the goal for improving road traffic safety.The specific work is as follows:1.In order to solve the problem of slow segmentation speed and low accuracy in the segmentation of lane line image,an improved PSPNet algorithm is proposed.The algorithm uses MobileNetv2 lightweight network to replace the main feature extraction network of the original PSPNet algorithm,which reduces the amount of network parameters and computational complexity;At the same time,hole convolution and feature fusion operations are introduced in MobileNetv2 network to strengthen the information fusion among feature layers,so as to avoid the problem that the segmentation accuracy will decline due to the large reduction of network parameters.The Caltech data set is used to verify the algorithm.The experimental results show that the improved PSPNet algorithm can accurately segment two types of lane lines: solid line and double solid line.Compared with the original algorithm,the loss value of the improved PSPNet algorithm is smaller,and the Miou value and MPA value reach 65.51% and 89.76% respectively.2.The DBSCAN algorithm with adaptive parameter adjustment is used to solve the problem of wrong clustering in the case of discontinuous lane lines.In order to avoid the single fitting algorithm can not fit a variety of linear lane lines,an adaptive fitting algorithm is proposed in this paper,which automatically selects the appropriate fitting line according to the deviation ? between the sample point and the fitting line.The experimental results show that the fitting algorithm can achieve accurate fitting of a variety of linear lane lines.3.Aiming at the problems of low accuracy and large number of parameters in M2 Det algorithm,an improved M2 Det algorithm is advanced.The improved algorithm uses MobileNetv1 lightweight network to displace the basic network in M2 Det algorithm,which reduces the amount of parameters and shortens the training time;The Basic RFB module is added in MobileNetv1 network,and the Mish activation function is used to substitute the Re LU activation function in the depth separable convolution to reduce the loss of information in the process of feature extraction and improve the recognition accuracy.The experimental results show that the improved M2 Det algorithm can accurately identify seven types of road traffic markings,and the recognition accuracy is better than other one-stage algorithms.Compared with the original algorithm,the m AP value is increased by 3.81%,the loss value is decreased by nearly 0.8,and the parameters are also greatly reduced.The visualization experiment results further show that the improved M2 Det algorithm can recognize more small objects and effectively avoid the phenomenon of missing detection.