Research On Atmospheric Visibility Measurement Method Basedon Automatic Image Recognition

The haze weather makes the external environment unclear,which has brought serious impact on people's daily life,especially in aerospace,transportation and other aspects.With the acceleration of the pace of life,the pressure of transportation has increased dramatically,it is necessary to accurately measure visibility.Therefore,the research of visibility measurement has been hot in recent years.In this paper,the digital camera is used to take the image,and then the target area is automatically selected for visibility measurement.The main research contents are as follows:(1)By understanding the development trend of atmospheric visibility measurement,analyzing the advantages and disadvantages of the existing visibility measurement methods,the digital camera method is finally used to calculate the visibility.In view of the disadvantage of Harris coner detection and Sift feature point detection operators,a Harris-Sift algorithm for fast image matching is proposed based on the characteristics of the two operators.(2)An algorithm for automatic selection of target area in image visibility measurement is proposed.After regular segmentation of the image,the OTSU Otsu method is used to segment the image by adaptive threshold,and then the image expansion is used to eliminate the complex areas,and the separated multi-connected areas are obtained.Then,the "pseudo" centroid of the expanded multi-connected region is marked,and then the "pseudo" centroid is used as the center to mark the target area of the size to be measured.Finally,using the transmittance formula of the dark channel prior algorithm,the target area where the largest point in the dark channel is located is selected as the final visibility measurement target area.(3)A visibility measurement system for automatic image recognition is established.Based on Cauchmide's daytime visibility measurement principle,dark channel prior algorithm is used to obtain atmospheric transmittance,calculate atmospheric extinction coefficient and retrieve atmospheric visibility.By comparing the images taken by different cameras in different directions for 10 consecutive days,this paper concludes that the image taken by the camera with higher pixels on the light-oriented side is the best and the measured data is more stable.Finally,the calculated 110-day image visibility is compared with the data of CJY-1G forward scatterer,and the accuracy is 92.30%.(4)A subsection visibility measurement model is established.Through the linear analysis of 110-day measurement data and CJY-1G data,the overall linear regression model was established,and the accuracy of calculation was 80.77%.After analyzing the reasons,a subsection visibility measurement model was established according to the actual situation,and the accuracy increased by 10.58%to 91.35%.