Research On Event Detection Technology For Highway Video Under Foggy Conditions

In the rapid development of Chinese highway,the proportion of traffic accidents and hidden dangers are increasing year by year.The major threats to the safety of the highway are two main factors:bad weather and traffic anomalies.Foggy has the greatest impact.Fog affects not only traffic safety,but also the contrast and color quality of the traffic video surveillance equipment degradation,and affects the application value of images.Vehicle retrograde and pedestrian intrusion also bring great hidden dangers.In this paper,we mainly focused on foggy video image clarification processing,pedestrian detection,retrograde detection.The paper based on the project of Jiangsu Provincial Department of Transportation which called "Intelligent Identification of Traffic Anomalies System Based on 3D Stereoscopic Vision'',the scientific research results can be captured by HD-SDI dome camera at K2090+700 of Nanjing-Hangzhou Expressway which was taken as the data source,and the video surveillance system of highway was established by using the established highway monitoring system.The detection of highway video events(pedestrian,vehicle retrograde)technology.The main contents are obtained:1.The clarity method of foggy traffic video image.A fog level detection method with contrast based on gray histogram was proposed.It was determined that the contrast limited histogram equalization in the HSI color space for the intensity by analyzing the R,G,B three-channel and gray-scale histogram characteristics of images under different weather conditions,and the results were evaluated.The method got a good effect in removing the light mist.2.The technology and algorithms in the video image preprocessing,including color image grayscale method,image filtering algorithm,image edge detection,threshold segmentation technology,and comparing the experimental results of different algorithms to determine the most suitable for experimental data.In order to avoid the noise in the image processing,and improve the processing speed,the detected region of the image was cut,then obtained the relevant ROI.3.The technology and algorithms of moving object extraction.The detection results of commonly used target detection algorithms are compared and analyzed,and it gets a clear background image by using the multi-frame median method;the paper improved the inter-frame difference algorithm,and a target detection algorithm based on the fusion of three-frame difference with the Prewitt edge detection operator was proposed,and got clear moving target.The morphological filtering method was used to remove the noise point in the image according to the noise patch area,and a relatively complete moving target was obtained.Connected domain labeling was applied to the extracted moving objects.4.The algorithm of pedestrian identification and retrograde detection was studied.According to the prior knowledge of pedestrian target feature,the pedestrian target was determined according to the target aspect ratio,rectangularity and complexity.The similarity function of the area,the displacement reliability function,the attribute similarity function and the histogram similarity function were defined,then the matching vehicle was determined by a comprehensive study of the similarity function of the moving target.It was used to judge whether the vehicle had retrograde behavior that whether or not the direction of displacement of the target vehicle was the same as the direction of traffic flow.Experiments showed that the method had a good effect on pedestrian and vehicle retrograde.