The present invention relates to an adaptive population counting method. An accumulation integration feature pedestrian detector is employed to perform pedestrian detection to obtain lots of detection candidate frames, the obtained candidate frames are subjected to preprocessing, the space, the time and the color feature of each candidate frame after preprocessing is extracted; based on the features, the Dirichlet mixture model and the Gibbs sampling theory are employed to perform clustering of the candidate frames; through the mode, population block mass areas are obtained; the population counting method based on angular points is employed to perform population preliminary counting of each population block mass area; and through the relation of the counting result and the feature points, the adaptive population counting method determines the population density, and aiming at different densities, different data fusion methods are selected to improve the accuracy of the counting result. The adaptive population counting method improves stability and accuracy of population counting detection.