Research On Image Defogging Algorithm Of Electronic Rearview Mirror System

The image acquisition of the electronic rearview mirror system will be disturbed by the haze weather,which will reduce the image quality and affect the driver’s subjective identification.Traditional defogging algorithm is used to improve image quality,but there are many problems such as high complexity and poor image processing effect.In this paper,aiming at the shortcomings of the existing research,starting from the atmospheric scattering model,the traditional defogging method and depth learning defogging method are used to study the image defogging algorithm.The main research contents are as follows:(1)Aiming at the problem that the parameter estimation of image defogging algorithm is insufficient,which leads to the incomplete and inefficient defogging,an improved image fusion fast defogging algorithm is proposed.Firstly,the image collected by the electronic rearview mirror system is downsampled,and two copies of the image are copied and converted to HSV color space respectively.For the first converted image,the V component is reduced;For the second image after conversion,the V component is processed adaptively and the S component is stretched linearly.Then,the first image is defogged by the improved dark channel algorithm,and the histogram is stretched.Finally,the two processed images are fused,resampled and output.Experimental results show that the algorithm has high efficiency of defogging,and effectively improves the image quality and contrast of electronic rearview mirror.(2)Aiming at the problems of insufficient transmittance estimation and insufficient feature extraction of the integrated network dehazing model(An All-in-One Network for Dehazing and Beyond,AOD-Net),an improved AOD-Net network model is proposed.Firstly,different scale convolution kernel is used to extract image feature information.Then,attention mechanism is used to allocate weight to extract image refinement feature and contour texture feature.Finally,the first two convolution operations of AOD-Net network are used to extract the extracted image feature twice.The network model is trained and joint parameters are estimated.According to the atmospheric scattering model,the joint parameters are estimated Image after fog.The experimental results show that the improved network model has stronger robustness,better subjective visual effect and less distortion after defogging;Objectively,the peak signal-to-noise ratio,structural similarity,ie entropy and standard deviation are greatly improved.(3)The above two improved algorithms are transplanted to the embedded platform to test the actual working effect.The electronic rearview mirror system transplanted with the improved AOD-Net network model is also installed on a commercial truck for testing.The test results show that the algorithm proposed in this paper can achieve better defogging effect even in the resource constrained embedded platform,and output images with high contrast and good visual effect,which proves that the algorithm is reliable and effective.