Common use of BACKGROUND EXTRACTION Clause in Contracts

BACKGROUND EXTRACTION. Typically, a computer vision-based detection system requires a background image that represents the base state of the area under observation. In the case of traffic detection, it is rarely possible to obtain an image of the observation area that does not contain any vehicles or other foreground objects. Therefore, it is necessary to extract the background image from the video stream itself. This is accomplished in an iterative fashion by using the pixels that make up an image. A grayscale image has only one value for each pixel that ranges from 0 and 255. A color image uses three color channels to represent a pixel’s color. These three channels in the RGB color space are the Red channel (R), the Green channel (G), and the Blue channel (B). Each channel has a value from 0 through 255 that represents the amount of that color. When the median background extraction algorithm is applied to a color image, the median value of each color channel needs to be found for each color pixel. The intensity (or luminance) of a color pixel is the value of grayscale converted from the R, G, and B color values by using Equation (5-1) (▇▇▇▇▇▇▇ and ▇▇▇▇▇▇▇▇, 2001). Grayscale = R * 0.30 + G * 0.59 + B * 0.11 (5-1) In this current study, the background image was obtained by constructing an image of the median value of each pixel from a collection of images: bgdi, j = median(img [n]) (5-2) i, j where: bgdi, j ▇▇▇▇, ▇ is the background image pixel value is an array of image pixel values In this study, we used a frame rate of 12 frames per second (fps) for video image processing. To extract the background image, 15 images spaced 20 frames apart were employed. By using the median value, it was assumed that the background was predominant in the image sequence. Figure 5-1 shows a snapshot of a video scene and the extracted background image for that scene. For data collections in locations with higher volumes (which would tend to obscure the background to a greater degree), a background extraction based on the mode of each pixel would be preferable (▇▇▇▇▇ et al., 2006). In high volume and congested situations where portions of the background are never visible, more advanced background estimations might be required. (a) A Snapshot of a Video Scene (b) Extracted Background