An hour of setup, a few hours of recording, and you have a defensible picture of what is actually happening on your street.

Most neighborhood speeding complaints stop at “I see cars going too fast.” That is hard for a council member or HOA board to act on. What works better is structured data: how many vehicles, at what speed, at which time of day, with the same answer holding up across multiple sessions. SpeedCam AI is built specifically for that job.

This guide walks through one good first study, end to end.

What you need

An iPhone 12 or newer running iOS 26. The on-device detector runs on the Apple Neural Engine; older phones cannot.
A stable view of the road. A window, a balcony rail, or a tripod all work.
Pro for recording your own footage. The Free tier already includes the full analytics dashboard with a bundled demo dataset, so you can explore before subscribing. See /pricing for the comparison.
About 30 minutes for setup, plus one to two hours per recording session.
At least three sessions across three days if you want a number you can defend in conversation.

Pick a vantage point

Stable mount matters more than expensive gear. A phone wedged on a window sill works fine.

A few rules of thumb:

Closer to perpendicular is better. Vehicles passing at roughly 90 degrees to the camera give the cleanest readings. Very oblique angles compress the image and make the auto-calibrator work harder.
Pick a location with at least 50 feet of clear road in view. The more frames each vehicle appears in, the smoother the speed reading.
Avoid pointing into direct sunlight (lens flare) and through dark tinted glass (color recognition gets less accurate).
If you have an exterior power source, use it. The detector running at 60 frames per second is steady on battery for a couple of hours, but a charging cable removes the worry.

Calibrate

There are two ways to calibrate, and they write to the same store, so you can use either or both.

Auto-calibrator. Open the Live tab and start a session. The app learns your camera angle in the background as traffic passes. The status pill in the upper area shows progress. The more vehicles, the faster it converges.

2-point reference line. Open the radial Tool Palette in the upper-left of the Live screen and pick the line tool. Tap to place two endpoints on a road feature of known length, then enter the distance. Useful real-world reference dimensions:

A crosswalk stripe is typically 6 feet wide
A standard lane marking is 4 feet long
A parking space stall is roughly 18 feet long
A typical sedan is about 15 feet long

The 2-point line is faster on quiet streets, where the auto-calibrator can take a while to gather enough samples. Anywhere with steady traffic, just let auto-calibration run.

Run a session

A few choices to make before you tap record:

Campaign tag. If you plan to compare locations or before-and-after conditions, add a tag (for example, “Oak St morning”). The Analytics tab can later filter to a single tag, and exports inherit the scope.
Speed limit threshold. In Settings → Capture, set the local posted limit and the over-threshold that triggers a saved incident. A common choice is 5 mph above the limit.
Calibration level. Off, Low, Standard, or High. Higher levels require more calibration samples before incidents get saved, which keeps the dataset cleaner. For a first session, Standard is a good default.
Focus filter. On the iPhone, go to Settings → Focus → SpeedCam AI. Toggling this filter keeps notifications from interrupting a session.

Recommended first session: 60 to 120 minutes during a peak window. Morning peak (7 to 9 am) and afternoon peak (4 to 6 pm) tend to surface the most consistent patterns.

Read the dashboard

Open the Analytics tab. Three numbers carry most of the weight:

85th percentile speed. The speed at or below which 85% of vehicles travel. Traffic engineers use this number to describe “what speed is the road actually being driven.” It is more honest than the mean because it ignores slow outliers.
Pace speed. The 10-mph range that contains the most vehicles. This is the dominant flow speed.
Compliance rate. Percentage of vehicles at or below the posted limit. The simplest answer to “is this above or below acceptable?”

Other charts worth looking at:

Speed timeline. A single-day view shows the scatter plus a rolling average. A multi-day view shows a 15th-to-85th percentile band: a wide band means inconsistent driving, a tight high band means everyone speeds.
Traffic by hour and 15-min volume. Confirms when the peak actually is, rather than when you think it is.
Risk heatmap. Speed against hour-of-day. Hotspots are easy to spot visually.
Vehicle types. Splits by car, truck, motorcycle, bus, bicycle, and pedestrian. A truck-heavy window can shift the conversation toward weight limits or commercial cut-through.

If you have an iPhone that supports Apple Intelligence, the on-device AI analyst can answer questions in plain English. “What was the fastest vehicle this week?” or “Plot traffic by hour on Tuesday.” Runs entirely on your phone, no cloud round-trip.

A PDF report is what most boards and councils want. Exports include the active session, summary stats, and burst stills for each over-limit capture.

For technical reviewers (city engineers, transportation consultants), JSON gives you the per-incident frame metadata, and TMAS matches the format used by traffic-engineering tooling.

Filters apply at export time: pick a campaign tag or a single session in the Incidents tab filter sheet, and the export inherits that scope. So if you want a “before” report from one campaign and an “after” report from another, run the export twice.

What this is good for

Trend detection. Tuesday morning is consistently 8 mph above the limit.
Comparative studies. This corner before and after a stop sign was added.
Documenting a complaint with structured data instead of impressions.
Pedestrian and cyclist counts. The detector classifies them separately.

What this is not

A radar gun. Per-vehicle readings have inherent uncertainty from the calibration source and viewing geometry. Treat individual incidents as imprecise; treat distributions across many vehicles as informative.

Court-admissible enforcement data. Laws around consumer-grade speed measurement vary widely by jurisdiction. SpeedCam AI is built for advocacy and trend documentation, not as a substitute for police-grade radar.

A way to identify specific drivers. License plates can be readable in burst stills. That is a sensitive category in most places. If you plan to share captures publicly, crop or blur identifying details first.

Local laws also vary on filming public roads, retaining recordings, and using data in formal proceedings. If you intend to share results publicly or present them to authorities, check the rules in your jurisdiction. SpeedCam AI does not provide legal advice.

Next steps

If your iPhone does not auto-detect a region for vehicle make and model recognition, see /support/vehicle-region.
If you want to push captures into Home Assistant, Zapier, or your own dashboard, see /automation for the webhook setup.
If you have a question that fits a single chart, the on-device AI analyst is faster than scrolling. Apple Intelligence is required.

How to run a neighborhood traffic study with your iPhone