Research On Nighttime Hazy Traffic Road Sign Detection And Recognition Technology Based On Image Processing

In recent years,the autonomous driving industry has received extensive attention,and a series of related technologies have also developed rapidly.In assisted driving technology,the detection and recognition of traffic road signs is an indispensable part.Traffic road signs can not only bring real-time road location information to autonomous vehicles,but also help update high-precision maps.However,to really apply this technology in practice,we must consider the impact of severe weather,such as low illumination at night and low visibility in haze.Therefore,around these issues,this article has launched the following research:（1）Night-time dehazing method for for traffic road signsDue to the problems of light source diffusion,color distortion and poor effect of the existing night-time dehazing method,this paper proposes a night-time dehazing method based on hybrid filter light estimation and transmittance optimization.Aiming at the inaccurate estimation of the ambient light at night,it is proposed to first perform side window filtering on the brightness image to ensure the illumination direction,and then use the guided filtering to refine the three channels as the local ambient light estimation;to solve the problem of light source diffusion caused by the application of dehazing at night,High-light area compensation is used to improve the transmission of the light source area.At the same time,for the uneven color and loss of details after dehazing,it is proposed to use guided filtering to correct the coarse transmission and then regularize the solution.Finally,the dehazing image is solved by the atmospheric scattering model.The experimental results show that the method in this paper can not only dehaze at night,but also dehaze during the day,which is better than existing methods both subjectively and objectively.（2）Traffic road sign detection methodFor the detection of traffic road signs,this paper designs a positioning method based on color and shape position.Transfer the to-be-detected image into HSV space to eliminate most of the interference information,and then locate the traffic road sign area through morphological processing and spatial location connectivity.Because there is a certain angle during acquisition,the image will be distorted.In this paper,the four corner points are detected by the Hough transform,and then corrected by the perspective matrix transform method.Finally,a standard-shaped traffic road sign was obtained.（3）Traffic road sign recognition methodConsidering that the technology in this article is mostly applied to moving vehicles,the collected pictures may have motion blur,and this problem needs to be solved.In this paper,based on the dark channel priori deblurring method,the bright channel is introduced to solve the problem of poor effect in the bright area at night.The brightness perception method combines the bright and dark channels well to estimate the intermediate latent image.Aiming at the problem of the lack of sparsity in fuzzy kernel estimation,the L1 norm is introduced into the L2 norm to perform joint constraints.Finally,the blurred traffic road sign image was successfully deblurred.（4）Traffic road sign recognition methodThe recognition algorithm of traffic signs is divided into two steps: Chinese character segmentation and recognition.In Chinese character segmentation,OSTU binarization is used to distinguish white characters from blue backgrounds,and interference items are eliminated according to the area of the connected domain,and then a single Chinese character is segmented through vertical and horizontal information analysis.Finally,in view of the problem of the Alex Net network model,optimization is performed by improving the activation function and the convolutional layer.After making the training set for training,the accuracy of Chinese character recognition is improved.