New Aggregation Strategies to Improve Velocity Estimation From Single Loop Detectors

Problem

Single loop detectors are commonly used to measure speed and flow on highways, but they have serious shortcomings in this area, as do conventional methods for estimating velocity from their information. For example, velocity and length cannot be measured independently at a single loop. Typically, operators set length to a constant value so they can estimate velocity. However, long vehicles can be up to four times as long as other vehicles, so they skew the data, and thus cause inaccurate results. Researchers have attempted to fix this by estimating mean velocity using aggregate flow and occupancy rates. We have come up with a new way to aggregate data, rather than manipulate aggregate data, to reduce estimation errors.

Method and Findings

We studied data from dual loop detectors, which can measure true vehicle velocity and length. By experimenting with the data, we found that median length estimates were much closer to accurate than mean length estimates. We also found the same to be true for the resulting velocity estimates: median estimates were much less subject to error.

Working with fixed time samples rather than the fixed number-of-vehicle samples we used in the preceding example, and insuring that the sample periods were long enough to be useful, we found similar results: there were significant errors in the mean velocity estimate and few in the median estimate.

We found that five minutes provided sufficient sample size. However, this may be too long in some cases, so we also set the time period shorter and sampled across five lanes at once, obtaining similar results.