At present, the flame recognition technology develops rapidly, which mostly uses the dynamic and static characteristics to identify and segment the flame. But the problems are that recognition error rate is still very high, and the high dependence of the flame recognition algorithm on hardware makes it difficult to apply to residential buildings. In this paper, a home fire alarm system based on Raspberry Pi and computer vision is designed to solve these problems, which takes the Raspberry Pi 3B development board as the hardware infrastructure and combines digital image processing technology. Firstly, traditional flame recognition algorithm was used to segment the flame area, and then SVM algorithm was used to identify and classify the flame, with an accuracy rate of 97%. Finally, a decision was made to decide whether to send warning information to residents. The system has strong expansibility and low cost, which can be widely used in residential buildings.