Video Smoke Detection System Based On DM6437

With the rapid development on the technology of digital video processing and computer network together with the hardware and software of the embedded system, the image processing technology and intelligent technology are used much more widely in fire detection system. There are more and more high-rise buildings and large space nowadays. The fast spreading fire in high-rise buildings and large spaces brings enormous difficulties when rescuing people together with properties and causes serious casualties and property losses. There has been an increasingly critical demand on real-time and accurate fire alarm in high-rise buildings and large spaces with stability and low false rate. The research and development of video-based smoke detection technology have become a trend nowadays.In this paper, video smoke detection algorithms is presented and analyzed. The algorithm firstly segments the areas of moving objects and updating the background,then decides whether the regions of interest is smoke or other objects. To develop the robustness of smoke detection technique, several natures have been combined: the energy variation in wavelet model, the color model, the structure similarity and the moving characteristics of the smoke. Finally, the fire alarm is decided according to the statistics of time and blocks.Then, the video-based smoke detection algorithm is running on DM6437 EVM made by TI. First, the drivers for video processing on both the video processing front end module and video processing backend module are prepared and used to capture images from camera and output processed images to display devices. Then, the video smoke detecting algorithm is implemented on the development board, DM6437 EVM, and the optimization is brought on both algorithm and DSP level. Optimizations on median filter and discrete wavelet transformation are applied on algorithm level while compiler options, the float-point calculation, the access to external memory, linear compilation and iteration are optimized to improve the algorithm efficiency and reduce storage space implementations which meet the demand of real-time with the consideration of the special structure of DSP.Experimental results show that the video-based smoke detection algorithm operates smoothly and performances well on smoke detection under different and complicated circumstances such as the smoke is partly occluded and people is moving in the detecting area. The algorithm meets the demand of real-time and robustness.