| Abstract [eng] |
In many computer vision systems it is important to classify parts of an image sequence as foreground or background. If it is possible to detect a foreground object further operations, such as recognition, identification or tracking, can be done on that object. Background subtraction is a particularly popular method to segment foreground and background. With this method the current image is compared with reference image of the background, and then the decision is made what is background and what is not by looking for changes at each pixel. In this thesis the adaptive background model calculation method is proposed. The key of the method is that the time of appearance of each pixel’s value is stored in memory and recalled later to update the background image used in subtraction operation to compute foreground objects. It is expected that this method will work well in ordinary image sequences where the foreground objects are the elements of urban scenery. The method probably will not work as well for objects which are of one color as the background because these pixels will be marked as background. |
