Lane Detection And Regional Vehicle Detection In Low-light Scenes At Night

In recent years,with the development of economy and the rapid increase of car ownership,traffic accidents frequently occur,and the problem of vehicle driving safety is becoming more and more serious.The rapid development of artificial intelligence technology and image processing technology has promoted the application of vision technology in intelligent vehicles.Lane line detection and vehicle detection are crucial parts of vision technology.However,in the low-light scene at night,the image obtained by the vehicle-mounted camera has low brightness,insufficient contrast,unclear image details and texture,and complex lighting,which causes great difficulties in the detection of lane lines and vehicles.From the perspective of improving the safety of intelligent vehicles in low-light conditions,this paper aims to improve the accuracy of lane lines and vehicle detection,and conduct in-depth research on lane line detection under low-light conditions at night and vehicle detection in the area.The main research contents of this paper are as follows:(1)Aiming at the problem of low overall brightness and insufficient contrast of night images,an image enhancement method based on fusion is proposed.The multi-scale retinex method is used to compensate the illumination of the image to improve the brightness of the image,and the two-dimensional empirical mode decomposition method is used to improve the contrast of the image.Finally,the illumination compensation image and the high-contrast image are fused to further improve the brightness and contrast of the image.(2)Aiming at the problem that the accuracy of lane line detection algorithm at night is greatly affected by ambient light and weather,a lane line detection method based on line segment intensity is proposed.First,the line segment intensity is used as the lane line weight,and the dynamic region of interest is obtained by estimating the road vanishing point,and the background interference information is filtered out;the density-based spatial clustering method is used to cluster the lane lines to eliminate false lane lines;The lane line sampling is discretized and extended to both sides at the same time to increase the number of data points during lane line fitting.The least squares method and random sampling consistency method are used to fit lane lines to improve the accuracy of lane line fitting.Experiments show that,compared with the existing methods,this method has a great improvement in the accuracy of lane line detection in low-light environment at night.(3)Aiming at the problem of low vehicle detection accuracy due to the lack of vehicle texture,edge contour and other feature information under low light conditions at night,a vehicle detection method based on the characteristics of always-on red taillights at night is proposed.The vehicle taillights are extracted from the Lab color space and HSV color space by using the dynamic threshold method,and the influence of other light sources in the environment is eliminated by logical operation and morphological filtering.The taillights are matched according to the vehicle taillight matching criteria,and the mismatch is eliminated by the similarity function,and the best matching result of the taillight pair is obtained.Test results in a variety of scenarios show that the method has high vehicle detection accuracy and robustness in low-light conditions at night.