In this research, a fire detection system has been built using the multi-feature fusion method. The detection starts with an improvised frame differencing method and a combination of Lab, RGB, and YCbCr color models to eliminate fire and nonfire objects in the frame. Then, feature extraction is performed by calculating the boundary disorder of the fire with the convex hull method, calculating the variability of the fire area, and calculating the stability of the centroid position of the object. Finally, the selected candidate fires are verified using support vector machine (SVM). For the test experiments, this study used 142 videos for the training set and 19 videos for the testing set. Each video has various scenarios such as various video resolutions and FPS, number of moving objects other than fire, environmental lighting conditions, angle, and where the video was taken. The test results show that the average accuracy of the system that has been built reaches 86.61%.
Keywords: fire detection, multi-feature fusion, support vector machine