Development And Research Of Driving Sight Distance Measurement System Based On Deep Learning

As one of the key parameters to be considered during road design,driving sight distance plays a crucial role in road alignment design and traffic safety.Relevant research shows that there is a negative correlation between the traffic accident rate and the size of the driving sight distance.The larger the sight distance,the significantly reduced traffic accident rate.Therefore,it is of great significance to detect the driving sight distance and detect sections with insufficient sight distance as soon as possible to ensure driving safety.The traditional detection method of driving sight distance is relatively cumbersome,requiring the acquisition of road geometric parameters and the calculation of sight distance through geometric models,which cannot achieve automated calculation.Therefore,this article converts the calculation of sight distance into the calculation of lane line length.Based on the accurate identification of lane lines,using the principle of visual distance measurement,a real-time driving sight distance measurement system is designed to improve the calculation efficiency and accuracy of driving sight distance.The main research contents are as follows:(1)Research on lane line detection algorithm based on deep learning.Although traditional lane line detection methods are mature,their generalization performance is poor,and they cannot adapt to complex and volatile road environments.Therefore,in order to accurately identify lane lines on complex roads,a lane line detection model based on deep learning algorithm is proposed.This model uses Res Net as the backbone network to obtain feature maps containing deep level lane semantic information;Using the designed anchors to pool the feature map and extract deep level lane line features;The pooled lane feature vectors are fed into the channel attention module SENet to allocate weight coefficients,improving the expression ability of the model;The weighted feature vectors are sent to two fully connected layers for classification and regression to complete lane line detection.Using the Tusimple dataset to verify the effectiveness of the model,the model in this paper has achieved good results in simple road scenes,while also being able to adapt to complex road scenes.(2)Establishment of a model for calculating the driving sight distance.On the basis of accurately obtaining the characteristic points of the lane line,a driving sight distance calculation model is established based on the principle of monocular vision ranging and geometric relationships;In order to solve the problem of changing the pitch angle of the vehicle mounted camera during driving,real-time calibration of the pitch angle is performed using the lane vanishing point to reduce calculation errors;Finally,the validity of the ranging model is verified through real vehicle experiments.(3)Development of a real-time ranging system.In order to achieve automatic sight distance detection,a real-time sight distance measurement system is developed using Python and Qt on the basis of accurately identifying lane lines and verifying the effectiveness of the ranging model.The system takes the real-time images collected by the vehicle mounted camera as input,outputs the visual distance calculation results in real-time,and displays a visual distance change map.In order to verify the effectiveness of the system,it is verified through real vehicle tests.