Personnel counting are a method of counting or accurately estimating the population in a given area. It plays an important role in many applications, such as public safety, crowd control and marketing analysis. The traditional personnel counting method based on video stream and electronic tag has high hardware cost, and the accuracy and reliability of the personnel counting method based on video stream are low under the condition of insufficient light or occlusion. This paper presents a method of personnel counting based on Wi-Fi perception. This method reconstructs channel state information (CSI) in Wi-Fi. CSI of multi-subcarriers effectively reduces the influence of multipath effect. CSI is reconstructed by deconvolution phase and linear transformation, so that phase information can be centralized in the form of clusters. It avoids the problem of too large range of original phase distribution and too high randomness. Based on Hampel filter, the singular data of carrier amplitude is removed, the interference of environmental noise factors on the number of personnel characteristics is reduced, and the accuracy and stability of personnel counting using wireless signals are guaranteed. Finally, the numbers of people are classified by SVM. The experimental results show that the counting accuracy of the proposed method is about 95.8%, which can accurately identify the number of people in indoor environment.