Research On Night Road Image Enhancement And Vehicle Detection Algorithms

In order to reduce the rate of road traffic accidents at night and provide clearer information about the road environment.It has been widely studied to obtain night road images based on vision and to enhance and detect them.Natural light in night road images is almost zero,and other light sources are interlaced and complicated,making the image brightness distribution uneven,image visibility and contrast decreased.At the same time,the vehicle outline information and texture information will also be blocked by the lights,resulting in low vehicle recognition,affecting the driver's timely response to road traffic conditions,and easily causing accidents.Therefore,the problems in the above-mentioned night road images are addressed.This paper designs a night road image enhancement algorithm to provide drivers with a clearer forward vision and road driving environment.A night road image vehicle detection method is designed to help the driver identify the road vehicles ahead and take timely avoidance measures to reduce the accident rate.The full text is divided into four parts as follow.Above all,the background significance of the subject is discussed.The status and progress of night road image enhancement and detection are introduced.At the same time,the problems of night road image enhancement algorithm and detection implementation are pointed out.Determine the research content of this article.Secondly,use CMOS cameras to collect night road images under different light source conditions,different time periods,and different locations.Establish classification data set and detection data set to provide data for subsequent experiments.Then,based on the Multi Scale Retinex(MSR)algorithm,this paper proposes an innovative optimized MSR algorithm for night road image enhancement.1)Convert the image from RGB color space to YUV color format.2)Construct the optimized MSR algorithm by using the inverse of the Just Noticeable Distortion(JND)as the coefficient before the MSR algorithm enters the image.This optimization algorithm is used to adaptively adjust the brightness of the Y channel of the image,and the U and V channels are adjusted proportionately.3)Combine the obtained image with the original image in a 1: 1 ratio to preserve image details.The limited contrast adaptive histogram equalization method is used to improve the image contrast,and finally an enhanced image is obtained.This paper focuses on the night road image enhancement algorithm,which alleviates the uneven brightness of the image,improves the image sharpness,and improves the image detail information.The classification algorithm is used to classify road vehicles.Compared with the original image,the enhancement image showed that the false detection rate decreased by 4.35% and the missed detection rate decreased by 2.61%.Finally,YOLOv3 convolutional neural network is used to detect vehicles in night road images..1)Set up a network experiment environment and use the GPU to train the network.2)Using Darknet-53 as the basic network,select 14 anchor boxes as the prior boxes,and save the training weights every 1000 rounds.3)A loss callback strategy is used during training to reduce model overfitting.This paper uses YOLOv3 network for night road image detection,which improves the detection speed while ensuring accuracy.It has significance for night road image detection.Using the detection data set for road vehicle detection,the detection accuracy rate is 93.66%,and the detection rate is 30 fps.