PTZ video visibility detection method based on luminance characteristic

摘要

Disclosed is a PTZ video visibility detection method based on luminance characteristic, which includes acquiring a road condition video image by utilizing a PTZ video camera, extracting the region of interest ROI of the road surface to obtain high constancy of selected pixels; acquiring precise road surface region by utilizing region-growing algorism based on Nagao filtering to ensure the illuminance constancy of the selected pixels in world coordinates; in the road surface region, extracting the contrast curve which reflects the luminance variation of the road surface, and searching the feature points of the luminance curve to calculate the human eye distinguishable and maximum far pixels in the image with an extinction coefficient; calculating the maximum visibility distance in combination with camera calibration to determine the visibility value. The present invention can take full advantage of existing PTZ camera to video the road condition and acquire the image without the need of providing any artificial marker. Monitoring can be in real time and has a low monitoring cost, and the monitoring requirement of large area road condition can be satisfied. Monitoring is stable and can not be disturbed by environment. It is a visibility detecting method with the advantages of simpleness, easy realization, high precision and excellent use effect.

主权项

1. A method to detect PTZ video image visibility based on luminance features of an image, with the characteristics of
using PTZ video cameras to obtain road condition video images, to extract the road surface domain of interest (ROI), and to achieve the selected pixel height consistency; using region-growing algorithm based on Nagao filter to obtain accurate road surface area, removing interference from the roadbed and vehicles, ensuring consistent illumination of selected image pixels in world coordinates; extracting contrast curve reflecting road surface brightness variation within the road surface area, identifying feature points of the brightness curve, and calculating the farthest image pixel distinguishable to human eye through the use of extinction coefficient; and calculating the maximum visibility distance in combination with the camera calibration, and determining visibility value, comprising the following steps: 1) collecting real-time road condition video images through existing PTZ video cameras; 2) determining the conversion relationship between the images and their corresponding world coordinates of the road surface through the camera calibration technology, after processing the video images from the video cameras, and calculating the distance of the road surface [ ] area of the video images and camera; 3) using Kluge model fit lane division lines to the projections of the video images, through the unknown parameters from randomly solved Hough model, designating the area between the lane division lines as current region of interest (ROI); limiting subsequent image processing to the ROI to ensure the consistency of image pixel height; 4) using region-growing algorithm, combined with ROI brightness criteria and adaptive Nagao filtering method, extracting road mask area accurately and focusing all subsequent processing on the mask area to reduce the number of calculations and to ensure consistent brightness of the selected pixels within the world coordinates;
calculating the gray scale median value median(Pg) of the bottom-most line of the ROI, selecting the pixel with brightness of median(Pg) as seeding point, scanning mask area according to bottom-to-top, left-to-right progressive scan principle, determining scanned target pixel P(i, j) whether it belongs to the road surface area according to the following growth criteria:41) Brightness balanceP(i, j) and median(Pg) satisfy
P(i,j)−median(Pg)≦ρnrminGmaxk (k=−1,0,1)  (6)In formula (6), ρ is a constant less than 1, nr is the number of separating rows between P(i, j) and initial seeding point Pg, Gmax refers to the brightness difference between the pixel and its top 3-neighborhood pixels, with top 3-neighborhood pixels referring to the three pixels on top-left, top, and top-right of the pixel; top-left brightness difference is Gmax−1, top brightness difference is Gmax0, and top-right brightness difference is Gmax1, among them:
Gmax−1=Gmax1∃mε{−1,i,i+1}Q(i,j)−Q(m,j+1)using the initial road surface domain with consistent illumination and consistent pixel height obtained in the previous step, coupled with the analysis on the trend of change in road surface pixel luminance caused by atmospheric extinction, identifying feature point of change, which is also the zero point for the second derivative of the luminance curve; 6) calculating visibility, including:
using the vanishing point coordinates obtained through camera calibration algorithm and the camera parameters, together with the zero point coordinates from the second derivative of luminance curve to determine atmospheric extinction coefficient;using the Koschmieder Theory then to deduce the relationship between atmospheric extinction coefficient and visibility, thus resulting in the visibility value.