Efficient pedestrian and bicycle traffic flow estimation combining mobile-sourced data with route choice prediction

Abstract

Accurate estimates of traffic flow measures like annual average daily traffic (AADT) are crucial for roadway planning, safety, maintenance, and operation. Due to resource constraints and high costs of traditional monitoring methods, we develop a methodology to estimate pedestrian and bicyclist traffic flows using mobile data sources, avoiding privacy issues of household surveys. The methodology is general, and potentially could be used with any reasonably comprehensive mobile source data set. To deal with erroneous and high variability data from mobile data sources we use different techniques to estimate and keep improving an origin–destination (OD) matrix constructed using the observed link flows to ultimately obtain reasonable approximations of actual link flows. We provide a non-linear optimization formulation along with a projected gradient descent based solution algorithm to solve this problem. Furthermore, we present the performance of the solution algorithm for several networks including the Twin Cities’ bicycle and pedestrian networks. We also compare the accuracy of our estimate with manually collected AADB and AADP counts from Minnesota Department of Transportation monitoring stations. For the Sioux-Falls network, the highest error from our model was less than 1%. These estimates can be improved by using existing methods to improve the quality of mobile sourced data as a pre-processing step to our method.