Microscopic Datasets: A Novel Approach Applied to Visualization of Spatiotemporal Flow Regions
Paul Nelson and Matthew J. Fields, Texas A&M University, May 2006, 33 pp. (167452-1)
A parsimonious approach to microscopic traffic flow datasets is suggested. This approach is based on provision of minimal data, along with a set of supporting tools termed as feature extraction operators that are intended to provide researchers with the flexibility to extract those particular features of the data that they desire to study. As an illustration of the possibilities of this approach, an attempt is made to validate the hypothesis that traffic flow decomposes into spatiotemporal regions representing one of the four classes of congested flow, shock wave, acceleration wave or free flow. This approach employs the classical conceptual framework of the field of pattern recognition, as applied to microscopic datasets. The specific microscopic dataset employed is that labeled “I-405 Northbound at Mulholland Drive, Los Angeles” in the 1985 study conducted by JHK Associates for the FHWA. The classical approach of plotting and manually analyzing vehicle trajectories is initially employed, to establish some approximation to ground truth. Then it is demonstrated that speed alone is inadequate to support the desired classification. Finally, a 4-means cluster analysis in velocity-acceleration feature space is employed to demonstrate that a spatiotemporal plot of the resulting cluster numbers provides a decomposition more-or-less as expected.
Keywords: Microscopic Data Analysis, Traffic Patterns, Classes of Traffic Flow, Pattern Recognition, K-Means Clustering
ENTIRE REPORT (Adobe Acrobat File – 3 MB)