Night Traffic Image Enhancement Based On Biological Vision Mechanism

At a time when autonomous driving technology is still immature,people are eager for advanced assisted driving technology to improve the comfort and safety of night driving.For a relatively dark driving environment,the on-board optical imaging sensor system cannot accurately perceive the surrounding environment information,so that the assisted driving system may make wrong decisions,thus bringing safety hazards to night driving.In fact,the nighttime traffic environment is a dynamic scene,and the lighting conditions in different scenes are different.To accurately capture the information in the driving scene is not only a test for the driver,but also a challenge for the advanced driving assistance system.Inspired by the different responses of photoreceptor cells on the biological retina to different light intensity stimuli and the different task division of the early visual dual pathways,the thesis adopts the idea of image classification enhancement.First,according to the brightness and local contrast characteristics of night traffic images,an image classifier is established,the night traffic images are divided into three categories,and then three image enhancement algorithms are proposed to enhance the three categories of night traffic images respectively,and the enhancement algorithm is evaluated by non-reference image assessment index.The results show that the enhancement algorithm has better performance.Finally,this paper develops a night traffic image enhancement software based on Matlab,which integrates night traffic image classification,enhancement algorithm and image quality evaluation index,in order to realize potential application value.The research content of the thesis includes the following three parts:In the first part,based on the brightness and local contrast characteristics of night traffic images,the thesis establishes a night traffic image classifier based on support vector machines,which can classify night traffic images into three categories,namely images with high scene brightness and scene brightness medium images and images with low scene brightness.In the second part,for the above three types of night traffic images,based on the physiological mechanism of biological retinal photoreceptor cells,the thesis constructs three image enhancement algorithms to achieve the enhancement of night traffic images.At the same time,three non-reference image quality evaluation methods（visual parameter measurement index,IL-NIQE index and quality index based on spatial spectral entropy）are used in this paper to quantitatively evaluate the image quality before and after enhancement.The experimental results show that,compared with the current mainstream dark light image enhancement algorithms,the three image enhancement algorithms designed in thesis have good performance in both qualitative and quantitative aspects.In the third part,this paper integrates the night traffic image classification task,image enhancement algorithm and image quality evaluation,and designs the night traffic image enhancement software,which is convenient for the display of image classification enhancement tasks.At the same time,the software also has certain potential application value in night traffic image processing.