Boarding and alighting passenger identification and passenger OD demand estimation using bus door videos

ObjectiveBus passenger origin-destination (OD) information serves as crucial data foundation for public transport operation and supervision. Existing studies primarily infer bus passenger OD demand based on station-level passenger flow combined with allocation algorithms. To overcome the limitations of traditional allocation algorithms, this study utilizes on-board video to identify individual passengers, thereby obtaining observed OD samples to reduce estimation errors.MethodThis study first employed YOLOv5, DeepSort, and instance segmentation technologies to achieve passenger tracking and noise-reduced feature extraction. Subsequently, it identified boarding and alighting passengers based on Bayesian probability model, as well as integrated an allocation algorithm to finally infer bus OD demand. To validate the effectiveness of the proposed method and its advantages in improving estimation accuracy, a case study was conducted on Shanghai Fengpu Express, comparing the results against two traditional methods.ResultThe proposed method significantly improves OD estimation accuracy. Compared with the method based solely on boarding passenger flow and the method based on both boarding and alighting passenger flows, the RMSE, MAPE, and MAE were reduced by approximately 84%, 78%, 58%, and by 57%, 52%, 34%, respectively. The findings not only enable the accurate acquisition of passenger OD distribution but also provide essential data support for bus operation and supervision.ConclusionThe findings demonstrate significant superiority and generalizability. While accurately capturing passenger OD distributions, this study provides vital data support for optimizing transit resource allocation, enhancing travel experiences, and facilitating operational supervision.