——
—
—
From camera counts to the whole park
The math, explained
The whole-park estimate starts from Taraval counts. When Taraval is unavailable, Ulloa fills the gap at ×1.2 on weekdays or ×1.4 on weekends and holidays.
- 1Choose the countKeep Taraval, use Ulloa when its measured count is higher, or adjust Ulloa when Taraval is unavailable.
- 2Estimate repeat viewsPeople may pass the camera more than once. Estimate repeats for walkers, runners, and cyclists so we don't overcount.
- 3Expand to the whole parkEstimate what share of all park visitors pass the camera.
Same camera data, your assumptions
Adjust the Estimate
Change assumptions below and watch the day and week estimates move. The camera counts stay the same.
For each activity, set how many camera views are repeats and what share of all park visitors pass the camera.
Why didn’t you count a particular person at a particular time?
Sometimes we miss them. Two people walking close together can be tracked as one, and someone hidden behind a group can disappear from the count. We also deliberately exclude people whose route looks like a crossing to or from the beach.
At the south-facing camera, we don’t count people on the separate lower path.
Do you count babies in strollers?
We don’t attempt to track strollers or count their passengers. We count the person pushing. So assume some extra little visitors beyond the hourly totals!
Are you counting people going to the beach?
If someone just crosses the promenade to get to the beach, we aim to leave them out. If they walk along the promenade and then head to the beach, they count as a park visitor too.
Anyone who spends at least two continuous minutes on the promenade counts as a parkgoer, even if they also go to the beach.
We use their path through the video to tell the difference. A crowd blocking the view can make that call wrong. The dashed boxes labeled “beachgoer (not counted)” show who we’ve excluded.
How accurate is this?
We're using computer imaging to track people. It's not perfect! People get lost, they get occluded by other people, they get counted when they shouldn't sometimes. Hopefully, it's pretty obvious that counts are directionally correct. You can slow the video down to 1x and watch for yourself! We likely double count some people, and we likely miss some people too. Hopefully it's a wash. And we're likely not off by an order of magnitude. Turning our count into whole-park attendance estimate adds two assumptions: how many views are repeats, and what share of park visitors pass our cameras at all.
Open the to change them and see how much the total moves. shows how the pieces fit together.
You misclassified a person. Why should I trust the estimate?
A walker labeled as a runner is still one person in the camera count. Getting their activity wrong doesn’t mean we missed them or counted them twice.
Missing someone, counting them twice, or excluding a park visitor as a beachgoer does change the person count. We do our best to minimize these!
How do you estimate times when the camera is too bright, too dark, or too blurry?
First we check the other camera. If Taraval is unusable but Ulloa is clear, we use Ulloa’s count multiplied by 1.2 on weekdays or 1.4 on weekends and holidays. When both cameras are usable, we take the higher count without a multiplier. We don’t add the two counts together.
Only gaps with no usable footage from either camera need the historical traffic pattern. That pattern follows the time of day, with separate curves for Monday–Thursday, Friday, and weekends. Usable counts from nearby hours adjust how busy the gap was likely to be, and we account for sunrise and sunset.
If ten minutes are lost to glare, we fill those ten minutes and keep the usable counts from the rest of the hour. If there’s no usable footage from either camera all day, we use an average for that type of day. A sudden rush during an uncovered gap can still be missed.