This paper raise an algorithm employing digital image process to detect fire smoke, it can promote the sensitivity of fire alarm system and decrease misinformation.This algorithm is composed of 6 parts: image pretreatment, background rebuild and moving object extraction, smoke detection by static features, smoke detection by dynamic features, smoke detection by color features, comprehensive judge.This paper discovered several efficient features of smoke, these features could be divided into static features, dynamic features and color features. Static features mainly include: (1) area become small from bottom to top (2) smoke can attenuate the high frequency signal of background (3) smoke shape is complex. Dynamic features mainly include: (1) inner' pixel value change continuously, and change frequencies successively decrease from bottom to top (2) the position of smoke source spot is immovable (3) shape of smoke change continuously. Color feature is that smoke can attenuate the color degree of background.This paper raised tow methods that how to divide the smoke region. Use conceptions such as "boundary chain code" and "round degree" to describe contour shape complexity. Divide the image into a large number of small blocks in analyzing high frequency attenuation and inner pixel change. Based on the division, the paper provides a way to describe the shape changing of moving objects. |