VLSI architecture for real time fire detection

Fire detection is an important task owing to the disaster that undetected fire may cause to life and property. For fire detection in real time video, the total processing time is critical. Parallel dedicated hardware can be used to solve this crucial problem effectively. In this paper we have proposed a parallel VLSI architecture for fast fire detection in a video sequence. After background subtraction, the algorithm has been categorized into parallel modules, where each module calculates one of the spatial or temporal properties of fire. The spatial properties considered here are color, saturation and intensity. The temporal properties are rate of change of the centroid position and change in the number of corners. The design is verified in XILINX VIRTEX IV FPGA with real time video. Evaluation has been done on videos containing flame as well as different flame-like objects. The worst case processing time requirement of the proposed method has been found out to be less than 4 ms for a PAL video frame (625 × 512), which is much less than inter frame temporal constraints of 40 ms for live PAL video.