An automatic flame detection system for outdoor areas

Abstract

Traditional fire detection depends on smoke sensors. This strategy, however, is unsuited for big and open buildings, as well as outdoor regions. As a result, based on computer vision systems, this research proposes an effective method for recognizing flames in open areas. To minimize data size without losing important information, integer Haar lifting wavelet transform is used to frame and analyze the input video. Then, three color spaces (binary, hue, saturation, value (HSV), and YCbC) are used in simultaneous color detection. In binary space, Otsu’s approach is utilized to determine automated intensity pixels. Additionally, using frame differences to reduce false alarms. According to the experimental results, the approach achieves 99% accuracy for offline videos and surpasses 93% accuracy for real-time videos while maintaining a lower level of complexity.