Multimodal Fire Alarm for Industrial Environments Using Ignition Sources

Small- and medium-sized industrial complexes that frequently use fire pits are continually exposed to fire hazards due to worker negligence or equipment malfunctions. In this paper, we propose a multimodal fire alarm system that simultaneously recognizes flames, fire pits, smoke, and worker movements to assess fire risk in real time and provide timely warnings. The proposed system incorporates a complex sensor module that includes an RGB optical camera, a thermal imaging sensor, a temperature and humidity sensor, a gas sensor, and a dust sensor. A deep learning model performs fusion analysis of RGB and thermal images to learn flame and heat patterns. Moreover, the system compares gas and dust sensor outputs to identify risk factors such as smoke, dust, and gas leaks. Further, it monitors worker behavior patterns in real time, enabling the early detection of careless approaches or dangerous behaviors around fire pits. A sensor fusion algorithm distinguishes between fire-induced and normal heat through RGB-thermal image comparison. Recurrent neural network filtering is employed to ensure time-series consistency, minimizing false positives and false negatives. Experimental results confirm that the proposed system maintained high detection accuracy and stability under various working environments and fire tool usage conditions. Thus, the proposed system successfully enables real-time fire risk assessment and alarm functions.