Lane Recognition Method In Tunnel Under Infrared And Visible Light Fusion

Due to the black hole effect in the tunnel entrance section,the uneven illumination in the middle section and the glare effect in the exit section,the features of tunnel lane line images taken by visible light cameras are easily lost.According to the radiation wavelength of the object,the infrared camera can recognize the road image collected by the visible camera under bad illumination conditions.Infrared and visible light fusion technology can supplement the information of infrared lane line images,and visible light video images can provide colour texture information to distinguish the road background and lane line.Firstly,the research status and progress of infrared and visible image fusion technology at home and abroad are introduced based on the analysis of transform domain,airspace and deep learning.Then the hybrid model was used in infrared and visible image fusion methods.Given the serious loss of feature association in the traditional sparse representation block processing mechanism and the local information supersaturation caused by the enhancement of different energies in infrared and visible light,this paper proposes an image fusion method based on the combination of sparse convolutional representation and local energy features.Secondly,taking the tunnel environment as the background,noise reduction and fusion processing are carried out for both infrared and visible lane line images to avoid the loss of visible and near-infrared information during fusion processing.Contrast-limited adaptive histogram equalization enhancement and guided filtering algorithms were used to preprocess the lane line images in infrared and visible tunnels.Non-subsampled contour waves were used to decompose the preprocessed images in multi-scale and multi-direction.The high and low-frequency information is improved,and then the high-frequency fusion coefficients are reconstructed to get the fusion image.The preliminary screening of lane line information was completed,and the curve determination was carried out.The Dynamic Region of interest method separated the target lane line area from the background.The improved Hough transform was used to recognize the straight lane line,and the least square method was used for curve fitting.Finally,the experiment shows that the CSR-SE-Energy fusion algorithm is better than the six typical fusion algorithms in index evaluation,the Improved Hough&LS MIo U’index compared with the Otsu algorithm is improved by 5.024%,AP’compared with the Otsu algorithm is improved by 4.488%.RT’has been reduced by 0.002s in Improved Hough&LS compared with RT.The comparison between T_n3 and SF-Energy-Q algorithm in the CSR-SE-Energy algorithm is reduced by 0.0553s,which is important in promoting the lane recognition process after fusion.When processing lane line image in the tunnel,compared with the other three models,σ_I’value of Improved Hough&LS is 7.310%,σ_P’andσ_T’’values of Improved Hough&LS are 2.601%and 0.020s²,respectively,higher than that of visible lane line imageσ_P andσ_T,with a small degree of dispersion.The overall operation cost is reduced,and the accuracy of lane recognition is improved.