This paper presents an image processing technique for automatically detecting obstructions on the road, such as stationary objects which prevent traffic flow. To utilize an image processing technique outdoors, it is most important that the technique be stable against lighting condition changes. The developed technique uses a reference background image updated following changes in lighting conditions and obstructions on the road are detected by comparing input images with the background image. In the monitored video image, changed areas are obtained by subtraction between an input and a background image, and between two successive frames. Then each changed area is classified into the following three groups: (a) moving object areas, (b) stationary object areas, and (c) areas changed due to lighting condition. The background image is updated only in areas (c) and those of no change. A cumulative image is obtained by accumulating a binary image composed of the changed areas (a) and (b) for a specified period of time. The stationary object is detected by thresholding the cumulative image. The present technique has been applied to the three different types of video images, and all stationary vehicles were identified accurately.
Abstract