Research On Intelligent Perception Technology Of Complex Vehicle-road Environment Based On Polarized Vision

As the effective technologies to improve road traffic safety,Automobile Intelligent traffic Advanced Driver Assistant System(ADAS)has attracted great attention from scholars at home and abroad.Among them,road visual perception technology is an indispensable part for vehicle-road environment research.Based on the current visual perception technology of vehicle-road environment,this paper researches the acquisition technology and fusion processing technology for complex vehicle-road visual images.Firstly,this paper based on the Monte Carlo simulation of the propagation process of light waves in a complex vehicle-road environment scattering media,a mathematical imaging model in the complex medium is constructed,the system imaging parameters are optimized,and the image enhancement acquisition scheme in a complex media environment is given.Secondly,according to the polarization characteristics of light waves,schemes for acquiring polarization information of complex vehicle-road environment were designed,a principle prototype system of polarization imaging was built and calibrated,correction methods for the field of view’s discrepancy and the inconsistent grayscale response of the system was studied,and the polarized image acquisition software was used to achieve accurate real-time acquisition of the polarized images of the vehicle-road.Then,by analyzing the collected Stokes parametric images,a multi-source image fusion algorithm was introduced.Aiming at the characteristics of complex vehicle-road images,an image enhancement fusion method based on polarization features was proposed,and it can realize enhancement of image detail information and its own polarization radiation characteristics.Finally,on the basis of the fused image,a complex vehicle-road environment multi-detail feature deep semantic classification recognition algorithm based on multi-scale pooling is proposed,which improves the effect of semantic classification and recognition on complex scenes,and realizes the correct classification of targets,such as roads,vehicles,and lane lines.It provides technical guarantee for the visual perception of safe driving assistance of vehicles in complex vehicle-road conditions.The real-time collected vehicle-road image is used to verify the algorithm proposed in this paper.The experimental results show that the recognition effect of semantic classification based on polarization features is better than the recognition effect of traditional visible light features.At the same time,the algorithm in this paper is superior to FCN-8s in terms of recognition accuracy and robustness,especially in some details recognition.Therefore,the application of polarized vision images to the identification and classfication of complex vehicle-road environmental targets can achieve accurate identification of vehicles,lanes,signs,and lane lines on the road,which is conducive to road traffic safety management.