The Study On Enhancement Method Of Pedestrian Detection Image In Tunnel Scene

Accurate detection of pedestrian targets based on video surveillance images is one of the key technologies of highway abnormal events detection.However,due to the enclosed tunnel environment,uneven illumination and serious light fog,the tunnel scene has low brightness and blurred imaging.At the same time,because pedestrian target is small,its pixel value is low and it has little difference from background,resulting in low accuracy of pedestrian detection in highway tunnel.Image enhancement technology can effectively enhance the quality of image,and reduce the miss detection of pedestrians caused by blurred picture quality and low pixel value of target and background.However,the existing image enhancement algorithms are mainly for daytime or nighttime opencast images,and its enhancement principle is not tenable for scenes between daytime and night like highway tunnel.Therefore,studying tunnel image enhancement algorithm has theoretical and practical significance.After analyzing the problems and difficulties of image enhancement under tunnel scene,this paper focuses on the construction of the tunnel imaging model and its parameter estimation,the algorithm of tunnel image enhancement based on adaptive illumination and the application of enhancement technology in pedestrian video detection.For a single tunnel image,firstly,referring to the traditional atmospheric scattering model and combining the characteristics of tunnel closure,a imaging model for tunnel scene is established.Then,three parameters of the imaging model,namely atmospheric light,transmission map and illumination are studied.In view of the atmospheric light parameters,the estimation method of global and local atmospheric light combination is adopted to improve the accuracy of atmospheric light parameter estimation.A fast estimation method based on image constraints is proposed for the parameters of the transmission.According to the parameters of the illumination,the difference of the statistical characteristics of the luminance components of the image is constructed and the C is constructed.The statistical feature recognizer of image brightness component based on mean clustering method realizes the classification and estimation of illumination in tunnel scene.Based on the above results,a fast and adaptive single image enhancement method for tunnel scene is formed.In order to solve the problem of time consuming and frequency selection forpedestrian video detection,the paper gives a local enhancement strategy for arbitrary pedestrian detection area.By establishing local region boundary model,the region is divided into several rectangular sub-regions,and each sub-region is enhanced.In order to reconstruct the restored image of local area,the application scope of the algorithm is enlarged.Combined the average time consumption of the single image enhancement algorithm with pedestrian detection,the approximate range of the video frame extraction frequency is calculated.On the basis of guaranteeing the effect of pedestrian detection,the optimum extraction frequency of video detection is optimized by the experiment method,so as to meet the detection requirement.Through a large number of tunnel image experiments,it is proved that the image enhancement of single tunnel proposed in this paper is superior to many traditional methods in the aspect of real time and objective evaluation of image quality.The application results show that the proposed tunnel image and video enhancement algorithm can effectively improve the effect of pedestrian detection and meet certain real-time performance.