Image fire detection based on color model and sparse representation

Some single feature information or their effective combinations which fire flame behaves are extracted as the basis of image fire flame recogniton in common algorithms. And large number of training samples are needed for learning procedure and parameters optimization. Moreover the recognition rate depends on the selection of features. Considering the global feature of fire flame, an algorithm was proposed based on color model and sparse representation for fire detection. Firstly, the regions with fire-like colors were roughly separated by color modeling in the space of HIS. Secondly, sparse representation model was built, and then the codebook of flames and suspected objects were constructed using PCA. Finally, the classification of fire flames and disturbances was implemented by calculating the minimum approximation residual error between testing samples and training samples using l₁-minimization. The experiment results show that the algorithm can effectively improve the classification precision and recognition speed, and also it achieves higher accuracy.