With the rapid development of Internet of things technology, people’s demand for adaptive non-contact sensing is growing. Aiming at the problems of traditional human perception technologies such as vision-based, poor sensor universality, and low accuracy, this paper proposes a new non-contact human presence perception technology based on WiFi, which sequentially preprocesses, extracts features, and classifies the collected channel state information, and establishes signal models corresponding to different states to realize human perception and recognition. Experiments show that the sensing accuracy of this method is high, up to 99 % , and it can be well applied in smart homes, medical care, and other life scenes.