Research On Monitoring Method Of Fog Visibility Based On Video

The visibility monitoring system in China mainly comprises of ground visibility observation.However,this method is expensive and has limited spatial resolution.China's highway image acquisition equipment is widely used,and it can reasonably use surveillance video to monitor visibility.Therefore,a method of measuring visibility in fog,based on monitoring video,is proposed in this paper.The images are divided into several panes,and then the calculated standard deviation and mean gradient of the pixel value in the pane is used as a feature.The features which were image feature matrix of optimal interest pane were extracted by correlation analysis.the function model is constructed by correcting the residuals of the linear model through the BP neural network model.Conclusions are as follows.(1)With the standard deviation and the average gradient as the monitoring factors,the brightness characteristics of the optimal interest pane are highly correlated with the visibility,indicating that brightness contrast is a key factor in visibility monitoring.Comparing the results of Donghai and Guannan,we find that the background of Guannan scene is dark trees.So the contrast between the target and the background is reduced,and the accuracy of the model is reduced.The Donghai scene is based on a light-colored sky.A strong contrast between the target and the background is more conducive to visibility monitoring.(2)When the visibility is less than 5000m,near 200m and 1500m is a segment sensitive area of visibility.The visibility interval is divided in this area,and the accuracy of the segmented model is further improved.The visibility interval of the image is determined by the random forest algorithm.The accuracy of the image classification is above 96%,which indicates that the random forest algorithm has strong ability to determine the visibility interval of the image.(3)It is proved that combination model that BP neural network modified linear residual has the advantages of high accuracy and small error.Comparing different light intensity,the accuracy of the combination model is relatively stable under the condition of sufficient light,with a relative error of under 10%.The accuracy of the model demonstrates general performance in low light intensity,with a relative error of about 15%.The combined model monitoring effect is better than a single linear regression model.The BP neural network model further improves the accuracy and universality of the model,and verifies that this method is feasible and effective in measuring the visibility in foggy days.