Many existing fire detectors are based on particle sampling, temperature sampling, relative humidity sampling and smoke analysis. Those sensors are applied very widely for low cost and simplicity, but those sensors must be installed in the close proximity of fire, otherwise fires can not be detected at all. That’s one of major drawbacks of traditional fire detection techniques. Thus those detectors can not be applied in large spaces or outdoors. And detectors often get fouled by dust and other non-fire particles due to direct contact with them. Some detectors make use of infrared sensors and that leads to high cost for video fire surveillance. To avoid fire disasters and reduce the cost of surveillance system, a lot of novel techniques have been proposed to improve the performance of fire detection. One of them is vision based fire detection. Vision based fire detection has many advantages over traditional methods. First, it has fast response to fires. Like many radiation based approaches, it detects fires as soon as they appear in sight. Second, it directly senses the location of fire. Third, vision based fire detection approaches not only detect fires far from detectors, but also provide additional information about fire, such as location, size, growing rates and spreading paths. Last, vision based fire detection systems can utilize hardware of widely available security surveillance to integrate both fire and security.