Estimating intersection turning volumes from actuated traffic signal information

Actuated traffic signals usually use loop detectors. The current practice in many cities is to install four consecutive loop detectors in each lane to reduce the chance of undetected vehicles. Due to practical reasons, all four loop detectors in each lane and other detectors referring to the same phase are spliced together. Thus, it is possible for several vehicles to be counted as one single car. This way of detector wiring to the cabinet reduces the accuracy of detectors for collecting traffic volumes. Our preliminary studies show cases with an error greater than 75 percent. Therefore, the purpose of this paper is to provide a simple method to obtain turning volumes from signal information in actuated non-coordinated traffic signals without using loop detector data. To produce the required data, a simulation was performed in VISSIM with different input volumes. To change turning volumes, a code was developed in COM interface. With this code, the inputs did not have to be changed manually. In addition, the COM code stored the outputs. Data were then exported to a single Excel file. Afterwards, regression and the adaptive neural fuzzy inference system (ANFIS) were used to build models to obtain turning volumes. The accuracy of models is defined in terms of mean absolute percent error (MAPE). Results of our two case studies show that during peak hours, there is a high correlation between actuated green time and volumes. This method does not need extensive data collection and is easy to be employed. The results also show that ANFIS produces more accurate models compared to regression.