<![CDATA[Fire is something that occurs as a result of a fire that is not handled properly and quickly. This incident is very dangerous considering the potential to cause material losses and loss of life. Various fire detection systems have been developed to detect and prevent fires. So far, many kinds of research on fire detection use fire and smoke sensors, and others, but that alone is not enough. It takes a tool that can detect a fire at the same time and know-how the conditions are when a fire occurs. With the development of technology, the use of computer vision is used to detect fires in real-time. Besides being able to detect fires, the system can also provide an overview of the current state of the fire through the camera. In this study, a computer vision-based fire detection was designed. Where it uses the You Only Look Once (YOLO) algorithm to detect fires and is accompanied by telegram notifications and buzzers. In the fire detection test, 4 different fire sources were used, namely small torch candles, large torches, and coconut fiber. From the test data for fire detection in various backgrounds, the accuracy value is 0.8, precision is 1 and recall is 0.8 during the day. While at night, the accuracy is 0.96, precision is 1 and recall is 0.96. The level of accuracy and recall values in testing various distances results in a fire detection system that will decrease as the distance between the fire and the camera is far away. The precision value produces a value of 1 at various distances during the day and night. That means the accuracy of the classification results is 100%. The stable precision value is influenced by fire detection readings, where the system never detects other objects as fire.]]>
