Flame Segmentation Characteristics of YCbCr Color Model Using Object Detection Technique

Abstract

The existing YCbCr color model for flame segmentation has a low segmentation performance for various colored flames and mis-segmentation for flame-like colored-object regions. An improved YCbCr color model using an object detection technique is proposed in this study to improve the flame segmentation performance of the existing YCbCr color model. YOLOv8, a deep learning model for object detection, was used to form a bounding box for the flame to prevent the segmentation of the flame-like colored-object region, and flame segmentation in the bounding box was performed. In addition, YCbCr rules were proposed to segment red and yellow flames to improve flame segmentation performance. The performance evaluation showed that the proposed model increased the intersection over union value by approximately 15.4% compared to that of the existing YCbCr model. In terms of the fire prediction performance evaluation, the precision, recall, and F1-score of the proposed model increased by approximately 15.9%, 28.2%, and 24.7%, respectively.