Fire Detection using Artificial Intelligence for Fire-Fighting Robots
Sreesruthi Ramasubramanian, Senthil Arumugam Muthukumaraswamy, A. Sasikala
Abstract
Fire-fighting robots are used in indoor environments to detect fires and extinguish them. Sensors such as flame sensors are currently used to detect fire in fire-fighting robots. The disadvantage of using sensors is that fire beyond a threshold distance cannot be detected. Using artificial intelligence techniques, fire can be detected in a wider range. Haar Cascade Classifier is a machine-learning algorithm that was initially used for object detection. The results obtained using Haar Cascade Classifier were not very accurate, especially when multiple fires had to be detected. Transfer learning from a pretrained YOLOv3 model was then used to train the model for fire detection to improve accuracy. The benefits and drawbacks of using deep learning for object detection over machine learning are highlighted. The algorithm used to obtain the target location the robot must move to use bounding box coordinates is also discussed in this paper.