Empirical Innovation of Computational Dual‐Loop Models for Identifying Vehicle Classifications against Varied Traffic Conditions

Clarifying traffic flow phases is a primary requisite for applying length‐based vehicle classifications with dual‐loop data under various traffic conditions. One challenge lies in identifying traffic phases using variables that could be directly calculated from the dual‐loop data. This article presents an innovative approach and associated algorithm for identifying traffic phases through a hybrid method that incorporates level of service method and K‐means clustering method. The “phase representative variables” are identified to represent traffic characteristics in the traffic flow phase identification algorithm. The traffic factors influencing the vehicle classification accuracy under non‐free traffic conditions are successfully identified using video‐based vehicular trajectory data, and the innovative length‐based vehicle classification models are then developed. The result of the concept‐of‐evidence test with use of sample data indicates that compared with the existing model, the accuracy of the estimated vehicle lengths is increased from 42% to 92% under synchronized and stop‐and‐go conditions. The results also foster a better understanding of the traffic stream characteristics and associated theories to lay out a good foundation for further development of relevant microscopic simulation models with other sensing traffic data sources.