Method of Obtaining OD Matrix for Bus Passenger Flow Based on Pedestrian Re-identification Algorithm

In public transport, the OD (origin-destination) matrix of each passenger's starting and ending points is of great importance for bus scheduling and network optimization. The difficulty is matching passengers getting off at the back door with those getting on at the front door in real-time. The leading technology used here is re-identification. The current re-identification method is mainly used in the security field of pedestrian re-identification, vehicle filling discrimination, and although the accuracy is high, the real-time performance is poor. In this paper, the classical method FastReID is improved by changing its attention module using GCNet to improve accuracy and by adding a fully connected layer to the BNNeck structure to classify other essential attributes of pedestrians while calculating the distance between the same passenger attributes; in the loss function, constraints are applied according to the distance between this attributes to reduce the intra-class distance. Finally, passenger features are stored in categories by attributes to improve the final retrieval efficiency. Experiments show that the method in this paper improves accuracy by 3.2 % and detection speed by 14.73% (69.58ms) per 100 retrievals compared to FastReID.