Video flame detection algorithm based On multi-feature fusion technique

Abstract

Video based fire detection of flame is potentially an applicable and promising technique in the field of pattern recognition. Over traditional methods, video based fire detection offers many advantages in large industrial applications. However, most current video flame detection algorithms on the features of color spectrum and spatial augmentation are always successful in extracting the flame region from environment in the image, while helpless against the potential interference sources, such as heat or light sources, motion resembling flame and moving operations of people and vehicles. For this reason, we put forward an intelligent video flame detection algorithm to distinguish flame from other lighting nuisances by a designed fire risk assessment model based on the analysis of static and dynamic characteristics of flame sharp angle and flame contour. The experimental results show that the solution has higher acceptance level of false/nuisance alarms with widely applied prospect.