Yuxuan Li¹ and Garret Washburn², ¹United World College South East Asia, Singapore ²California State Polytechnic University, USA

House fires in Singapore are quite common due to the widespread lack of consumer fire prevention and safety systems installed in residential areas [3]. This paper proposes a solution, the ScorchVision system, that uses a machine learning model and a Raspberry Pi with camera to detect fires in camera view and alert users through the accompanying mobile application. The whole system consists of a Raspberry Pi hardware configuration with an on- board custom trained machine learning model using PyTorch, a mobile application written in Dart using the Flutter frameworking, and a back-end server created with the Flask framework that communicates directly with a Firebase database [2]. Throughout development there were a few major challenges that required troubleshooting, all of which had to do with working with the individual technologies as there are a lot of moving parts within the system. To ensure the system works as expected, two dif erent experiments were performed to find the accuracy and reliability of the machine learning models classifications as well as the average updating time from the hardware to the database [4]. Both experiment results, included in this paper, were quite positive in displaying the reliability of the ScorchVision system. Overall, the ScorchVision system is a promising new way for homeowners to take charge of keeping their home safe from fires in a fashion that is much cheaper than other posed solutions on the market. Additionally, the system is entirely open source and free to download and use, enabling privacy with complete ownership over the system and

Fire Detection, Machine Learning, Artificial Intelligence, Home Safety, Mobile Application