Method Of Active Infrared Smoke Video Detection Based On Fusion Texture Feature

Video fire detection is widely used in the field of fire protection due to its advantage of high efficiency, embedded type, non-contact detection and visualization. However, ordinary video surveillance system cannot work well in the low illumination environment, which could lead to the failure of fire detection. Flame and smoke are produced when a fire occurs, and smoke usually appears before the flame in the early stage. The video smoke detection technology can find the fire in such key time and provides more precious time for evacuation and firefighting. Therefore, this paper presents a systematical study on the infrared video and image of early fire smoke in the low illumination environment by the active infrared video camera technology. The active infrared video smoke detection algorithm is also developed with aim to provide a theoretical support for the fire detection and monitoring in the low illumination environment. The main research contents and conclusions are listed as follows:An acquisition system of active infrared camera is designed to capture the fire smoke in the low illumination environment, with aim to build the sample base of fire smoke video. The active infrared camera, armed with infrared light supplement lamp, is used to collect the experimental data of early fire smoke. The early fire smoke experiment considers the influencing factors, including the distance between camera and smoke, fuel type, smoke concentration, wind interference, human interference and light interference.Smoke texture characteristics in low illumination environment are studied by combining the smoke characteristics of Local binary patterns (LBP) with gray level co-occurrence matrix (GLCM). And then the recognition method based on the combined characteristics is improved. After pretreatment such as Median filtering, three frame difference method is used to test the smoke movement area. Finally smoke texture feature of the suspected area combined LBP with GLCM is extracted to identify of smoke and non-smoke.At last, in order to verify the validity of the active infrared video smoke detection algorithm in the low illumination environment, this paper analyses the experiment results. The results show that the fusion texture method is always effective, no matter the change of distance, material, smoke density, wind, and light and pedestrians interference. Furthermore, the fusion texture method can be used to distinguish smoke and interference. The ROC curve result shows that the fusion texture characteristic holds high robustness, in comparison with the single LBP as well as traditional fire characteristics. So the method of active infrared smoke detection based on this feature can be applied to identify smoke in low illumination environment.