Fusing Nonmotorized Traffic Data: A Decision Fusion Framework
Abstract
This dissertation explored an uncharted research territory, fusion of nonmotorized traffic data for estimating reliable nonmotorized demand measures. The research was divided into three sequential stages. The first stage involved developing and applying a guideline to process and homogenize available nonmotorized data sources to estimate demand within a specified scope. Multiple data sources were utilized to develop five bike demand models to estimate annual average daily bike volume at intersections in Austin. Following an in-depth discussion of both nonmotorized traffic data and fusion characteristics, the second stage proposed a decision fusion framework divided into two broad categories: fusion without benchmark data and fusion with benchmark data. Under the first category, four fusion algorithms, including a novel approach, were investigated. Under the second category, a robust state-of-the-art statistical tool, the Dempster Shafer (DST) method, was endorsed. Dempster Shafer with credibility context, proposed by this study, offered a unique way to incorporate subjective judgment of the experts in the mathematical fusion formulation. The third stage was focused on applying the fusion framework on both actual and simulated data to demonstrate the efficacy of the fusion algorithms. The findings illustrated that the novel weighted voting fusion generated a fused estimate of comparable accuracy to the best source estimate when applied to four demand sources. However, when five source fusion was conducted, the accuracy decreased. When applied to simulated data of multiple scenarios, the DST method outperformed the individual source estimate in most cases. Based on the data, categorization and discounts, the change in accuracy varied from -10% to 7%. Therefore, fusion exhibited the risk of obtaining worse-off results. Moreover, the proposed DST approach outperformed the traditional approach in most cases (above 80%), underscoring the merit of incorporating subjective judgment. The fusion considering knowledge and context is expected to contribute to the field of decision fusion. While the framework offered an additional option of analysis, it is up to the analyst or practitioner to consider and decide the course of option in adopting fusion endeavors given the trade-off between effort and the change in confidence, coverage, and accuracy of the outcomes.
Subject
FusionExposure
nonmotorized activity
demand models
Crowdsourced data
Dempster Shafer
Decision Fusion
Citation
Sirajum Munira (2021). Fusing Nonmotorized Traffic Data: A Decision Fusion Framework. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /195406.