Capacity Estimation of Signalized Road Links from Crowd-sourced Trajectories
Sampled positions from vehicles acting as mobile sensors are already an essential source of traffic state information for drivers, fleets, and traffic operators, and knowledge from traffic theory and practice would add value to such data source to be used in traffic planning and engineering. Saturation flow rate (SFR), which is an indicator of signal capacity, is among the important parameters used in signal planning. However, direct field measurements of this parameter are limited in space and in time, and current methods require context and geometric information to be applied in adjustment functions, which typically fail to account for the effects of various influencing factors. Moreover, no reliable method exists for measuring time-varying saturation flow rate for movements constrained by other higher ranked movements. Therefore, the aims of this work were to build and test numerical and data processing methods for estimating signalized intersections’ capacities from sampled trajectories and thereby advance data-driven traffic planning. An initial investigation of one signalized straight movement combined a field measurement technique of the SFR with adaptations of existing methods for estimating signal settings in order to assess the suitability of using trajectories in signal capacity estimation. This investigation identified parameters that require calibration, the most important of which was the average vehicle length (equivalently, the jam density). Subsequent research therefore investigated the role of vehicle properties and their correlations, while also seeking a more reliable reference measurement than what can be obtained from the formulae in professional guidelines. These two challenges were addressed through investigations that included measurements from processed videos. These enabled analysis of the correlation between vehicle size and signal crossing speed, which in turn enabled the use of speed as an indicator of the average vehicle size in the queue. Thus, the SFR specific to the cycle of the observed vehicle can be estimated. After the mentioned investigations, the final proposed approach is presented then experimented on a complete intersection as follows. First, a clustering-based method for determining signal settings is described and tested on all twelve movements of the intersection. Second, measurements taken from aggregated trajectories (rather than aggregated measurements from single trajectories) are used to construct a triangular fundamental diagram, from which the SFR is estimated. Results from the different movements are compared and interpreted with respect to their influencing factors, and a sensitivity analysis on the sample size is presented. In sum, the proposed method offers several contributions that advance the state-of-the-art use of trajectories for signal planning, and can be applied to develop data-driven methods of managing unpredictable traffic flows and composition.