A comparative study of different color spaces in computer-vision-based flame detection

Color information plays an important role in computer-vision-based fire/flame detection. The purpose of this study is to determine the most effective color space by performing an objective comparison among 18 different color spaces in terms of classification accuracy and class separability measures. In the comparison of classification accuracy, a bag-of-features (BoF) method is proposed in the paper. The experiments are based on 2000 images, including interfering objects collected from the Internet. According to the experiment results, the proposed BoF method can greatly improve classification accuracy for positive samples compared with alternative algorithms, while also ensuring the accurate classification of negative samples. The sRGB and PJF color spaces perform more effectively according to the experiment results. A trade-off can be found in the two color spaces in the classification accuracy of positive and negative samples; the best classification accuracy of all samples is achieved in the PJF color space with the J-F plane. In a comparison of class separability measures, the first three optimal values are also achieved by the sRGB and PJF color spaces. Therefore, the two color spaces are recommended in computer-vision-based flame detection systems according to our experiment results.