Passerby · the footfall and traffic research tool
How many shoppers, kids and cars pass your store?
Passerby turns your cameras into a research tool: pedestrians, cyclists and vehicles, counted by direction and by the quarter hour. Try it below on a live Caltrans camera or a video of your own. Or count any YouTube video from a link. For the London version, which counts 10-second clips, click here.
Pick a camera on the map, find one by road or place, or try a video of your own.
Amber dots have live video, where everything passing is counted. Grey dots show a still image every few minutes.
Loading the cameras…
Live video Still images only
Who it’s for
Retail site selection
Compare candidate locations by the people and cars that actually pass them, at the hours your customers shop, before you sign a lease.
Landlords and leasing
Show prospective tenants real foot and car traffic for a storefront, by day and hour, instead of a guess.
Outdoor advertising
Count the audience passing a billboard or screen, split by pedestrians, cyclists and vehicles, and by direction of travel.
Events and planning
Measure how an event, a road change or a new bike lane changes traffic, with the same camera before and after.
What Passerby does
- Works with the cameras you have. Add any camera that sends a video stream (most IP cameras do), or upload a recording. Each one becomes a counting point.
- Counting lines where they matter. Draw one across the sidewalk for pedestrians, one across the road for vehicles, one at a store entrance.
- Reports by the quarter hour. Pedestrians (adults and children separately, on a street-level camera), bicycles, cars, trucks, buses and motorcycles, each by direction. Compare locations, days and weeks, and export everything as a spreadsheet.
- No faces, no plates, no recordings. Only counts are kept. The video is looked at frame by frame and not stored.
How the counting works
Found in your browser
An object detection model (EfficientDet-Lite2 through Google’s MediaPipe, about 12 MB, downloaded once from this site) looks at each video frame on your own device and marks every pedestrian, bicycle, car, truck, bus and motorcycle. No video goes to AI Web Center, including a video of your own. The faster the device, the smoother the boxes follow the traffic.
Followed frame to frame
Each one is linked from one frame to the next by where it was and where it was heading. When its centre crosses the counting line it’s counted once, in the direction it crossed, with the time. Draw your own line across the sidewalk or the lanes you want to count.
What makes it miss
Highway cameras are high and far away, so pedestrians and cyclists are only a few pixels tall; a camera at street level counts them far better. People crossing a street walk along the counting line rather than across it, so draw a pedestrian line across the crosswalk to count them. This demo counts all pedestrians as one group: telling adults and children apart needs a street-level camera, such as one in a shopping area. Night, rain, glare, crowds and a camera that’s turned or zoomed lower the count, and cars and pickups get mixed up. Treat these counts as estimates.
Camera list as of 2026-10-26, from Caltrans’ public data at cwwp2.dot.ca.gov. Images and video load straight from Caltrans. Map © OpenStreetMap contributors.