Estimating Average Annual Daily Pedestrian Volumes at Intersections based on Turning Movement Counts

Abstract

There is a focus on increasing the use of active transportation and, consequently, a need to have pedestrian traffic volumes such as Annual Average Daily Pedestrian Traffic (AADPT) for infrastructure planning and safety analysis. Traditional methods rely on the deployment of dedicated sensors to count pedestrians, but this limits the number of locations at which counts can be obtained and therefore does not permit estimation of AADPT for all intersections in the urban area. The focus of this thesis is to propose and evaluate methods for addressing this limitation. The proposed methods assume that (i) dedicated sensors that provide continuous pedestrian volume counts are deployed at a small number of intersections within the urban area, and (ii) 8-hour turning movement counts (TMCs) are available for intersections for which AADPT are to be estimated. These two assumptions are normally met in practice. Within this context, the problem of estimating AADPT can be divided into five sub-problems, namely: 1. Calculating AADPT with missing counts in a dataset 2. Selecting and implementing a set of count data filters 3. Associating specific continuous count sites with each other 4. Finding suitable factors groups for short-term count sites 5. Converting short-term counts to AADPT estimates This thesis examines the existing methods in the literature for solving each of these sub-problems and proposes several extensions. By solving all the subproblems, there is a hope that reliable average daily estimates from pedestrian data collected alongside turning movement counts can be obtained. It is recommended to use the AASHTO method for determining continuous count site AADPT values or solving sub-problem 1. For the data filters, it was determined that using pre-exiting filters from the literature with some adjustments was appropriate. However, a new null count filter was needed for the dataset. For grouping specific continuous count sites, existing solutions from the literature were incorporated into this work along with a proposed k-means clustering approach. Specific land uses and temporal metrics were incorporated into linear regression models for the purposes of predicting specific temporal trends and placing a short-term count site in a factor group. Lastly, the AADPT estimation methods were all taken from the literature and are mathematically adjusted to handle 8hr to 24hr conversions. The methods are applied to a set of field data from Milton, Ontario and Pima County, Arizona. The results indicate that the AADPT estimation error metrics still are much larger for count sites located within 1km of a high school and, consequently, a modified factor grouping method is proposed for sub-problems 3 and 4.