Fire detection has evolved with a long history. Those traditional heat and smoke detectors utilize, respectively, the heat and smoke produced by fire to react. While the flame detectors absorb IR/UV emitted from the flames and trigger if it gets to a certain level of intensity. Most of the above mentioned detectors are just elements of a complicated electrical alarm system. They can not be installed to operate by their own. Moreover, they tend to react late or erroneously by environmental interference because of the limitations to their operational principles. Most fires create a great amount of smoke in the initial stage. When it pervades all over the space, people can then be aware of the situation from a distance. Recently the cost of computer vision products has been decreasing in a great deal, which brings many applications in this field. It has been an obvious stream to use CCDs to directly detect the visible light of fires and smoke. This research will develop smoke detection software by a graphical programming language, which grabs the images of a workplace with smoke emitted. By transforming RGB color system to HSI system and filtering noises by changing pixels, the software is able to recognize the smoke foregrounds in each video frame. Next, the analysis of the intensity distribution of the foregrounds is performed subsequently. The manual finding of the features is to further eliminate the noises that are left during the previous filtration. Finally, the largest smoke foreground in the filtered image will be locked up by a tracking frame. Therefore, the software can accurately determine the real smoke.