Rangefinder
Semantic Source Selection
An iOS rangefinder that selects the best depth source per frame across six heterogeneous sensors—from LiDAR at arm's length to terrain ray-casting at 2,000 meters.
Every depth sensor has a range where it works and a range where it lies. LiDAR is centimeter-accurate at 3 meters but returns nothing at 50. Neural depth estimation looks plausible at 30 meters but hallucinates mountains into nearby hills at 200. No weighted average of sources that individually produce wrong answers can produce a right answer.
Rangefinder replaces weighted-average fusion with a deterministic priority chain. Instead of blending all sources, the system selects one authoritative source per frame—the one that is mathematically trustworthy at the current distance and scene geometry. A second hypothesis is tracked independently via dual Kalman filters, giving the operator context about what's in front of and behind the primary target.
| Source | Range | Method |
|---|---|---|
| LiDAR | 0.3–12m | Direct time-of-flight depth from ARKit |
| Neural Depth | 2–50m | DepthAnythingV2 with continuous LiDAR calibration |
| Geometric | 5–200m | Ground-plane trigonometry via IMU pitch |
| DEM Ray-Cast | 20–2,000m | Digital elevation model terrain intersection |
| Object Detection | 20–1,000m | Known-size object pinhole ranging via CoreML |
| Stadiametric | 5–1,500m | User-directed bracket overlay with target presets |
Dual Kalman Filters
Foreground and background depth tracked independently with automatic filter reset on semantic source switches.
Neural Hard Cap
Inverse-depth noise amplification makes neural estimates unreliable beyond 50m. The system knows when to stop trusting the model.
Operator Guidance
IMU-based coaching engine modeled on military marksmanship doctrine: stability detection, respiratory pause windows, reading lock confirmation.
Golf Mode
Stadiametric ranging with USGA regulation flagstick preset (2.13m). Club-selection accuracy: ±2 yards at 150 yards at 5–8× zoom.
DEM Map Verification
MapKit picture-in-picture showing the ray-cast hit point on satellite imagery. See exactly where the terrain intersection landed.
Ballistic Solver
Inclination-corrected distance output with optional holdover computation for angled shots. True horizontal distance, not line-of-sight.
Layer 8 UI Presentation
RangefinderView, SceneRangeOverlay, HUD, Operator Guidance
Layer 7 Operator Guidance Engine
IMU stability, coaching hints, respiratory pause capture
Layer 6 Output Processing
Ballistics solver, inclination corrector
Layer 5 Temporal Filtering
Dual Kalman (fg + bg), motion-aware smoother, outlier reject
Layer 4 Semantic Source Selection
Priority state machine: one source per frame
Layer 3 Depth Source Estimation
LiDAR, Neural+Cal, Geometric, DEM, Object, Stadiametric
Layer 2 Sensor Abstraction
CameraManager, InclinationManager, LocationManager
Layer 1 Hardware
LiDAR, Camera, IMU, GPS, Magnetometer, Barometer