Counting a Minnesota winter, one skier at a time.

Three Rivers counts carefully. Every summer, staff and volunteers run two-hour counts at more than 550 park and trail entrances, alongside vehicle counters, trail counters, sales data and visitor surveys.^[5] Those counts feed the Metropolitan Council’s annual estimate, which put Three Rivers at 14.7 million visits in 2025.^[7]

But the counts happen between Memorial Day and Labor Day. Winter, spring and fall visits are calculated from the summer numbers using survey multipliers.^[6] For a park whose winter is built around skiing, that leaves the busiest season for some trails as an estimate. Three Rivers wanted to see what direct winter counts could show: total use, the share of each activity, and how hard each entrance is used.

Waypoint placed 8 solar cellular cameras across the park: four at park entrances, two on dedicated ski trails and two at crossings where trails meet. Each was strapped to a tree about ten feet up from a ladder, with no power, trenching or permanent hardware. A Three Rivers tag on each camera and signs at the trailheads told visitors a trail study was under way.

Every passage is classified by activity, such as skiing, walking, walking a dog or riding a fat bike, faces are blurred automatically, and each count links to the photo behind it.

In 28 days the eight cameras recorded 22,514 passages. 72% were on foot and 26% on skis; bicyclists, vehicles, snowmobiles and animals made up the rest. Among pedestrians, 6% walked a dog and 8% were running, and 27% of people were in a group. The busiest site, Site 4 near the center of the park, carried 9,646 passages, more than the five quietest sites combined.

Use built from mid-morning and held steady from 10 a.m. to 5 p.m., peaking between 4 and 5 p.m. with 2,263 passages across the pilot, as people came out after work and school.

On Saturday, February 21, Three Rivers’ Winter Play Day offered free cross-country ski passes and rentals at Hyland.^[4] The park recorded about 2,400 passages that day, the busiest of the pilot and roughly twice a typical day, most of them at Site 4. Because each count has a photo behind it, the park can report the festival on its own and keep it out of everyday averages.

In the first week of March, highs reached the 40s and 50s and the trails softened. At Site 6, on the ski trails, skiing almost disappeared: on several days the camera counted only a few dozen people, nearly all on foot. When snow returned mid-month, so did the skiers. On Saturday, March 14, the park’s busiest two hours of the week saw about 450 passages, more than half of them on skis.^[8]

Weather sits beside each day’s numbers in the dashboard, so Three Rivers can see how much of a winter season depends on snow and grooming, not just on demand. That is something a multiplier applied to summer counts can’t show.

A park-wide total hides how different each trail is. Site 6 watches a trail with groomed ski tracks; Site 1 watches a trail with no ski tracks. At Site 6, 46% of passages were skiers. At Site 1, there were none: 94% pedestrians, plus a few bicyclists and vehicles.

Because the cameras classify every passage by activity, Three Rivers can see the share of skiing, walking and biking at each entrance and trail, not just how many people went by. That is what grooming, signage and winter trail-use rules are decided on.

Winter Play Day exposed a problem. Children learning to ski stop, turn around and linger in front of a camera, which triggers it again and again. Early counts treated each photo as new people, so a few families could look like a crowd.

From this data Waypoint built a repeat-subject filter. It compares each detection with the ones in the photos just before it and marks a person already counted as a duplicate instead of a new visitor. It cut false counts from repeat subjects from 31% to 3%. See how noise filtering is validated.

From March 3 to 14, a Waypoint camera ran beside an Eco-Counter MULTI infrared counter on a narrow, four-foot trail, through dry ground and deep snow, 20°F to 80°F, day and night, with walkers, skiers, cyclists, dog walkers and animals.

Day by day the two counters rose and fell together: their daily totals correlated with an R² of 0.95. But the camera counted about half as many as the infrared counter. Without manual counts there was no way to tell whether the infrared counter was overcounting or the camera undercounting. That open question led to the next study, with MnDOT on the Nine Mile Creek trail bridge, where a better-placed camera came within 7–12% of an infrared counter.