A Fire Source Localization Algorithm Based on Temperature and Smoke Sensor Data Fusion

Traditional video surveillance, temperature-based or smoke-based fire source location methods are difficult to timely and accurately locate the fire source in warehouses with the characteristics of burning intensely, smoke spreading quickly, and being sheltered by shelves and goods. To overcome the drawbacks, a deep-learning-based fire source localization algorithm with temperature and smoke sensor data fusion according to the different stages of the combustion process is proposed in this paper. The temperature and smoke concentration information are collected from sensors distributed in different spatial locations of a warehouse. A convolutional neural network is used to exact the fusion data feature. The deep learning algorithm is adopted to construct the fire source localization model where the fusion data feature of temperature and smoke concentrations are the inputs and the fire source coordinates are the outputs. By using Fire Dynamics Simulator, a warehouse that meets the practical application is constructed and kinds of fire scenes are simulated. The experimental results show that the RMSE of the model localization reaches 0.63, 0.08, and 0.17 in three stages respectively, which verifies the effectiveness of the proposed fire source localization algorithm.