The accurate prediction of visitor flow in tourist attractions presents a significant challenge within the tourism industry and holds substantial reference value for both park management and tourist experiences. Addressing this, our study develops a predictive model specifically tailored to tourist sites using trajectory data. Recognizing the limitations of current algorithms in identifying accurate stay regions, we utilize a segmentation method predicated on change points. This approach integrates a Back Propagation (BP) neural network with the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering algorithm to enhance the precision of stay region identification. Building upon this foundation, we further incorporate Gaussian fitting techniques to construct a comprehensive crowd prediction model for tourist attractions. The research results verify that the model in this paper can estimate the passenger flow better by predicting the passenger flow of Zhongshan Park in city A. It is found that when the passenger flow is below 15000, the passenger flow is less. When the passenger flow is larger in the range of 15000~30000, and when the passenger flow is more than 30000, it will be saturated and crowded, and the model constructed in this paper has a more accurate passenger flow. The model built in this paper has a high accuracy of people flow prediction value.