Research On Vehicle And Lane Case Segmentation Algorithm In Intelligent Vehicle Assistance System

Road safety is a growing problem,with thousands of drivers and passengers dying every year in our country.In the past ten years,researchers have carried out a lot of research work,hoping to improve the safety,save lives and reduce the number of deaths on the road by monitoring the road environment.In this paper,the key technologies in intelligent vehicle assistance systems ar e vehicle detection,lane detection,lane departure,and vehicle ranging.At first,t he Mask R-CNN instance segmentation detection algorithm is optimized to imple ment vehicle detection.Secondly,the two branches instance segmentation detectio n algorithm is optimized to implement lane detection.Finally,the vehicle and la ne detection results are used to predict lane departures and perform distance mea surement.The specific research content is as follows:(1)Based on the Mask R-CNN model,the algorithm is improved.At first,t he Res NeXt-101-FPN network is used to replace the Resnet residual feature pyra mid network during feature extraction to improve the accuracy of small target ve hicles while reducing the amount of calculation.Secondly,the use of switchable normalization instead of batch normalization solves the problem that the accuracy rate is affected by batches.The improved algorithm has a 5% higher accuracy ra te than the original algorithm in the detection of small targets.(2)Introduced the commonly used target detection algorithms CNN,R-CNN,Mask R-CNN and two branches instance segmentation of convolutional neural ne tworks,and compared and analyzed their performance.The choice is more robust in multi-target detection.The two branches instance segmentation model is used f or lane detection.Based on the two branches instance segmentation model,the n etwork structure is optimized,and the Focal Loss function is used to train the n eural network in LaneNet lane detection to solve the problem of loss instability caused by the imbalance of positive and negative samples.The traditional SGD optimizer is used solve the efficiency problem during the experiment.The impro ved algorithm loss was reduced from the original 0.0244 to 0.0158.(3)Based on the research of vehicle and lane detection,research the driving safety early warning model in intelligent vehicle assistance system,including lane deviation model and distance measurement model.A model based on vehicles cr ossing lane boundaries that takes into account changes in moving vehicles is use d.This model uses information fusion of yaw angles and pixel distances to stud y lane departures.Coordinate transformation and two-point distance formulas are used to achieve effective vehicle ranging.