Big data is a part of the traffic data collection landscape, but bigger doesn’t always mean better
In light of changing traffic patterns during the pandemic, some U.S. states have halted traditional data collection programs for the time being and are instead pushing for the adoption of big data powered software platforms. Big data platforms, however, require specific usage guidelines when it comes to supplementing traffic count programs (see, Virginia’s Report), because big data isn’t an appropriate tool for all use-cases. In a research report published by Virginia’s DOT, it not only includes what big data is typically being used for (i.e., origin-destination studies and updating trip tables), but it also provides some measures of big data count accuracy (e.g., accuracy related to volume-based thresholds and rates of error based on sampling). With the accuracy measures provided, the report also suggests appropriate use-cases for big data and where data produced by other technologies (e.g., video-based counts) are likely a better fit.
What’s apparent, in light of the report, is that not all data is created equally. The qualities of data are distinct based on technologies employed to collect the data itself. Video-based traffic counts, for example, are in fact complementary to big data counts and are becoming more important in the world of big data, due to the high-fidelity potential of video-based counts. While big data’s unique value is to provide a macro-level overview of a traffic network, video-based counts are uniquely positioned to provide micro-level insights, as well as visually-derived insights (e.g., qualitative insights such as lane occupancy, gap analysis, near-miss).
Examples of what location, video-based data does well:
Turning movement counts (TMC), lane-based automatic traffic recordings (ATR), pedestrian and cyclist pathway and junction studies, modeling calibration
Examples of what big data does well: Origin-destination (OD) for trip tables, trip purpose, annual average daily traffic (AADT) estimates.
When you need high-fidelity, verifiable, and accurate data, video-based counts provide the best, most granular data.
- Any counts that require multi-modal data, from demand studies to complete street assessments. Read about the City of Chicago, for example.
- Safety studies where qualitative data is as beneficial as quantitative data, and where visually derived information contributes to the study
- Time-sensitive studies, where the accuracy of the volumes at specific times and specific locations is significant
- Locations where the vehicles traveling on roadways fall below 1000 vehicles per hour
Ultimately, the question of “which data is better?” depends on the use-case and the need for precision within a study. If the specific volumes and/or movements at specific times are less relevant to a study, macro-level big data is an option. If specific volumes and/or movements at specific times are important, then you need ground-truth data.
“It’s a question of what accuracy and precision you need to answer the questions you have. Big data methods provide value in aggregate (both temporal and spatial), but as you need to dig deeper into details, ground-truth data is required.”
Sajad Shiravi, P. Eng
If you’re interested in learning more about how to collect ground-truth data and the appropriate use-cases, whether it’s to power specific studies or to calibrate your big data models, reach out to the team of experts at Miovision.
