Abstract: In order to solve the problems of low detection accuracy and single application scenario in current urban rail transit train passenger flow analysis, this paper designed an intelligent passenger flow analysis system for urban rail transit trains based on heterogeneous ensemble learning method. The system was based on cloud edge collaboration architecture and adopted the grouping Voting method to integrate YOLOv5s (You Only Look Once v5s), FCHD (Full Convolutional Head Detector), and CSRNet (Network for Congested Scene Recognition) models as the base models, ultimately implemented functions such as passenger flow statistics, congestion analysis, and auxiliary passenger clearance. The experiment was conducted using monitoring image data of a train on a certain line of Beijing urban rail transit. The experimental results show that the model used in the system has better detection performance compared to other basic models, which effectively improves detection accuracy, and enriches applicable detection scenarios.