Here is some practical advice:
- HEADCOUNT SNAPSHOTS: Collect three snapshot counts, or more, with different occupancy levels. You may have to visit at different times of day, or even different days, to get ground truth data that are:
- in the bottom 20% of expected occupancy
- in the middle 20% of expected occupancy
- at the maximum 20% of expected occupancy
- THRESHOLD COUNTS: Collect three threshold counts, or more. Each threshold count interval should be at least 15 minutes long, with aggregate gains or losses in occupancy are better.
- Counts where detection zone occupancy remains static are not as useful as counts where the detection zone occupancy grows or declines significantly.
- Counts are better when the two ground truth counts have different levels of traffic: one low-traffic, one high-traffic.
- None of the threshold counts should be close to zero. If the threshold count is lower than 10, keep counting beyond the 15 minutes period.
- OVERALL: The more headcount snapshots and threshold counts that are collected, the more accurate will be the calibration.
- Remember that if all thresholds are not concurrently counted, the threshold count is of zero value and cannot be used to run a calibration job.
- It is not unreasonable to have a half-dozen or dozen ground truth counts (snapshot or threshold) per calibration job. A count with five thresholds represents a single count.
- Rejecting counts of questionable quality will deliver a better calibration outcome. Reject these counts at the time of collection rather than waiting until calibration time.
- Read our article about accuracy calculation.
Article is closed for comments.