Extracting traffic patterns from loop detector data using multiple changepoints detection

摘要

In this paper, we first introduce the Pruned Exact Linear Time (PELT) (6)—a segmentationapproach for detecting multiple changepoints—to automatically identify the onset of congestedperiods of freeway operation using original, disaggregated, 30-second loop detector occupancydata. The purpose of the algorithm is to detect and map phase transitions in the occupancy data,keeping the general features of the traffic pattern while substantially reducing time incomputation, retrieving, and presenting data with computation complexity that is only O(n). Byusing PELT, the start and end of the congestion period is identified automatically. The algorithmis tested on data from over 1000 mainline detectors in Orange County, California, USA both fora single day and for a month. The compression ratio of occupancy data is about 38.5, allowing anopportunity to analyze and monitor traffic in a more efficient way. This research provides anapproach to quantify and display both the beginning of the congestion as well as total congestionduration on temporal-spatial maps that could lead to an inexpensive means to improve the qualityof ramp metering settings and real time traffic monitoring.