Demographics

Machine estimates of apparent age, visible skin lightness and vehicle value from sampled, face-blurred camera views. Not census data; no identity or race is inferred.

No imagery, tracks or camera positions are published here. How camera video is handled

Area

San FranciscoRolling 7-day window

Connecting to recent estimates…

PEOPLE

Apparent age

Years

—Waiting for estimates

Approximate visual age, with each estimate's uncertainty. Log-normal curve; ages stay above zero.

VISIBLE COLOR

Apparent skin lightness

CIE L* 0–100

—Waiting for estimates

Darker (0) to lighter (100) lightness of visible skin, including uncertain visual guesses. Lighting and camera color affect it.

VEHICLES

Estimated market value

U.S. dollars · log-normal

—Waiting for estimates

Log-normal approximation of AI vehicle-price estimates, with a longer high-price tail. Broad visual guesses are refined by checked asking prices when available; this is not an appraisal.

NEIGHBORHOOD COMPARISON

Apparent age by neighborhood

By name · mean ± 1σ · n per row

Neighborhoods with fewer than three active cameras, or fewer than 20 estimates, are withheld, as are estimates from fewer than three cameras and any figure that would let a withheld one be worked out from the citywide figure.

NEIGHBORHOOD COMPARISON

Apparent skin lightness by neighborhood

Darker 0 → lighter 100 · CIE L* · n per row

The same lightness scale as the curve above. Lighting, shadows and camera white balance affect every estimate.

APPARENT PRESENTATIONAverage apparent-female probability (machine estimate)

For each sampled person the model gives a probability between 0 and 1 that their presentation looks female, from visible physical cues only. This is the average of those probabilities. It is not a share of people, does not establish anyone's gender identity, and sits near 50% whenever cues are weak. Most estimates are low confidence. No interval is shown.

Collecting estimates for the neighborhood comparison.

How to read these estimates

Every figure is a machine estimate from a sample of face-blurred camera views over a rolling 7-day window, or since the day shown where the samples start later: not census data and not a count of residents. Sampling rotates through active cameras and avoids analyzing the same tracked object twice; a camera's estimates stay in the window while it is offline. Overlapping cameras, occlusion and reconnects may still repeat a person or vehicle. Each row shows its sample size, n.

Public figures follow fixed rules. A distribution from fewer than 20 estimates reads "Not enough samples", and one whose estimates come from fewer than three cameras reads "Fewer than 3 cameras"; any neighborhood with fewer than three active cameras is withheld entirely and reads "Fewer than 3 active cameras". Where the citywide figure minus the published neighborhoods would reveal a withheld figure, the published neighborhood with the fewest estimates is withheld as well and reads "Withheld to protect another area". Cameras whose location is still being verified are grouped as "Location pending". Individual images, camera rows and exact sample times are never published; times are rounded down to the hour.

Age and vehicle value use log-normal approximations, which assign no probability to negative values. Vehicle value uses a linear dollar axis so the longer high-price tail stays visible. The curve is matched to the published mean and variance, not an observed price histogram or proof that real prices follow this distribution. Skin lightness uses a Gaussian on the CIE L* scale from 0 (darker) to 100 (lighter). Each curve combines variation between estimates with each estimate's own uncertainty; low-confidence guesses carry broader uncertainty and are included. Estimates with no visible support are excluded and counted in the coverage line.

Faces remain blurred. Skin lightness is a best visual estimate of visible skin color, refined by a measured exposed-limb patch when one is usable; faces are never sampled. It does not infer race, ethnicity, health or Fitzpatrick type. Apparent presentation is estimated only from visible physical cues and does not establish gender identity.

Vehicle value starts with a broad visual estimate from design, category and likely age. When two current comparable listings can be checked against their source pages, those prices refine it. The spread includes uncertain identity, age, mileage and condition.