Outdoors on a clear day, Wheria behaves like any parked-car application: a pin on the map, a distance readout, perhaps a photo of the parking spot. Underground, the same receiver that gave you five-meter accuracy at street level now reports horizontal error exceeding thirty to eighty meters — or loses fix entirely, leaving the map drawing a stale position from minutes ago. Indoor mode is not an attempt to make GPS work better through software tricks; it is pedestrian dead reckoning, the classical technique of estimating position by integrating discrete steps and heading changes from an inertial measurement unit, augmented by barometer floor hints and honest uncertainty display.

When you tap Park, Wheria captures an anchor: the last usable GPS fix if one exists, with latitude, longitude, and reported accuracy radius, plus a timestamp and optional photograph of a pillar number or row marker. If you are still at street level, the barometer records a pressure baseline at the garage entrance. The application then starts a high-rate IMU session — CoreMotion on iOS, SensorManager at game or fastest rates on Android — and begins drawing a polyline representing every step you take after leaving the car. Each detected footfall advances position by your stride length in the direction of current heading, computed in a local east-north-up coordinate frame centered on the park point. Heading itself comes from gyroscope integration between steps, corrected by the magnetometer when the magnetic field appears trustworthy, a condition that fails frequently in steel garages and is treated with appropriate skepticism as described in our compass calibration post.

What each sensor contributes — and what it cannot

The accelerometer and gyroscope together enable step detection by band-pass filtering acceleration magnitude around walking cadence, roughly one point five to three hertz for typical walking speeds. Peak picking with an adaptive threshold identifies individual footfalls; the gyroscope confirms turn events between steps. A common student mistake is to double-integrate acceleration to obtain position, which drifts meters per second because consumer MEMS biases are milligram-level and integration squares time. We do not use raw acceleration for position — only for step events.

The magnetometer measures Earth's magnetic field vector, providing absolute heading when the local environment has not distorted the field with steel columns, rebar, or the ferrous body of a parked car. In clean outdoor conditions it works well; in parking structures we down-weight compass measurements and rely on gyroscope integration between steps, surfacing an explicit "compass unreliable" indicator rather than pretending north is known. The barometer reports absolute atmospheric pressure, and because pressure decreases roughly twelve pascals per meter of altitude gain near sea level — a relationship derived from hydrostatic equilibrium — the change in pressure since parking can suggest which basement level you reached, though elevators and weather fronts complicate the reading as we detail in barometer floor math. GPS remains active until horizontal dilution of precision or reported accuracy exceeds usable thresholds, at which point we hide the drifting dot rather than display a lie.

Stride length and the error budget

Default stride length is approximately zero point seven two meters, adjustable from height using the anthropometric estimate of zero point four one five times height in meters. Users may optionally calibrate by walking a known twenty-meter distance in a straight line while the application counts steps and stores a personal stride coefficient. Stride error dominates long walks: a five percent overestimate on stride produces four meters of longitudinal error after eighty meters walked, before heading uncertainty adds a lateral component. That is why Wheria displays distance bands — "about forty-five meters" rather than "forty-four point seven meters" — and colors a confidence chip green, yellow, or red based on fusion covariance, connecting to the Extended Kalman Filter treatment in our Kalman filter post.

Honest interface choices

The path polyline ages visually: older segments fade so you recognize a stale trace if you paused mid-garage for coffee. When heading uncertainty exceeds twenty-five degrees — common near EV charging stations where DC cables and transformers distort magnetic fields — Find mode suppresses turn-by-turn voice guidance and shifts to distance-first presentation: walk approximately sixty meters, then rescan compass in a clearer area. We would rather stay quiet than shout wrong directions into a concrete wall. ktuyen validates these behaviors across garage architectures in her test matrix: open slab structures like Landmark 81, narrow spiral ramps in mall basements, outdoor-to-indoor transitions, and the magnetometer nightmares around electric vehicle bays. Success means the user finds the car within one wrong-turn correction on at least seventy percent of walks under one hundred twenty meters, using phone sensors alone without SLAM maps or Bluetooth beacons. For the step detection algorithm underneath, see step detection and IMU odometry; for why GPS fails in the first place, read why GPS lies to you.