A Fire Detection System On Videos Based On CUDA

In recent years, with the development and application of image recognition technology, fire detection which is based on image recognition technology has made some achievements in the actual needs and the promotion of academic research. In this paper, with the understanding of fire characteristics in video sequence, the system uses the technology of video detection to detect and analysis the characteristics of flame and smoke in videos. For flame detection, system uses Gaussian model to remove background, color model to detect fire and Fourier descriptors to detect the dynamic characteristics of fire. In smoke detection, system uses wavelet transform to analyze smoke's features which include transparency and jitter. The system gets energy sub-graph from wavelet transform of the flame and detects the characters with energy changes and jitter. In order to calculate the integration of the various characteristics of information, the system uses neural networks to process characters from the three modules and gets the probability of fire. This paper also uses CPU-GPU co-processing architecture to parallel the image processing algorithms. And with the help of parallel, system improves the performance and the throughput of the system. The system could work with good stability and accuracy, and maintains fast processing efficiency.