In this paper, through the research of digital twin technology, combined with the application of vision sensor, artificial intelligence chip and deep learning algorithm technology, the real-time monitoring and alarm system of elderly fall and abnormal posture based on digital twin technology is developed. The system collects the data of the posture and behavior of the elderly, and then presents them in the cloud by digital mapping after the calculation and analysis of artificial intelligence chip. Once the safety threshold is deviated, the alarm can be activated to avoid or mitigate the injury caused by the fall of the elderly. Through product validation and trial operation of Tianbao Nursing Home in Hongkou District of Shanghai, the user can set alarm thresholds in different time periods and regions, thus achieving the preset purpose of the product.