GPS provides 3–5 meters accuracy. For AR experiences in an urban environment, this is catastrophically insufficient: overlaying navigation arrows on a specific building entrance or showing an AR annotation above a particular sculpture in a museum courtyard with 5-meter error is impossible. We implement Visual Positioning System (VPS) for mobile AR applications turnkey — from scanning the space to SDK integration. With VPS, the user points the camera at the surroundings, the algorithm matches the frame against a pre-recorded visual map, and returns a position with 10–30 centimeter accuracy. We will evaluate your project and select the optimal solution.
How VPS determines position
Two stages: mapping (offline) and localization (online, real-time).
| Stage | Time | Resources |
|---|---|---|
| Mapping | 30 minutes to 2 hours | Phone with camera, SfM server |
| Localization | 200–500 ms | Server or device |
Mapping. A room or outdoor area is scanned: an operator with a phone or specialized rig walks the entire area, recording video. Keyframes are extracted from the video, a Structure from Motion (SfM) algorithm runs — building a sparse point cloud and a set of 6DOF camera poses. Feature point descriptors (ORB, SuperPoint + SuperGlue for better accuracy) are indexed in a database for fast retrieval.
Localization. The phone captures a frame, sends it to the server (or processes locally on powerful devices). An image retrieval algorithm finds the nearest keyframes from the database → PnP (Perspective-n-Point) computes the camera pose → returns a 6DOF transform. All within 200–500 ms with server-side processing. For offline localization on the device, frames are recorded into a local database, increasing privacy.
How to build a visual map for VPS
- Shoot the space with at least 60% frame overlap, uniform lighting, and full coverage of user paths.
- Upload the video to the Immersal or other provider's server.
- SfM processing: 30–120 minutes, after which you receive a mapId.
- Accuracy verification: test localization at control points.
If the interior changes — rescan the changed areas and merge with the existing map via the provider's console. We guarantee that the final map will provide stable localization even with partial camera occlusion.
Which SDK to choose for VPS
| Provider | Coverage | Offline | Accuracy | Use Case |
|---|---|---|---|---|
| Google ARCore Geospatial API | Cities with Street View | No | ~10–30 cm | Urban AR experiences |
| Immersal SDK | Custom maps | Yes (device) | ~2–5 cm | Indoor spaces |
| Niantic Lightship VPS | Lightship wayspots | No | ~10–20 cm | Mobile AR games |
| Apple ARKit + GPS | Outdoor, iOS only | Partial | ~1–3 m | Quick prototype |
| Microsoft Azure Spatial Anchors | Custom | No | ~1–5 cm | Azure integration |
For custom indoor spaces (museum, office, warehouse), Immersal is best — it provides 10x better accuracy indoors than Apple ARKit + GPS. For urban AR experiences on iOS and Android — ARCore Geospatial API or Niantic VPS. Azure Spatial Anchors is good when integrating with the client's Azure infrastructure.
How to integrate Immersal into a native application
Immersal provides a REST API for cloud localization and a Unity SDK — the latter we are not interested in; we work natively.
iOS: HTTP request to https://api.immersal.com/localize with JPEG frame and camera intrinsics → JSON response with pose in the map coordinate system → convert to ARKit world space via matrix transformation.
struct LocalizeRequest: Encodable { let token: String let fx, fy, ox, oy: Double // camera intrinsics from ARCamera.intrinsics let image: String // base64 JPEG } // Get mapToWorld matrix, apply to ARSession.currentFrame Android: similarly via retrofit2 + moshi, camera intrinsics from CameraCharacteristics.
Do not run localization on every frame (200 ms latency) — trigger on position change of 1+ meter or when ARCore tracking is lost.
What's included in the work
- Technical specification with provider selection and architecture.
- Scanning and building visual map for indoor spaces.
- SDK integration into native iOS and/or Android application.
- API documentation and map update instructions.
- Accuracy and stability testing on target devices.
- One month post-launch support.
Timelines and cost
Integration of ARCore Geospatial for urban AR experience: from 2 to 4 weeks. Custom VPS with Immersal for indoor space including scanning: from 4 to 8 weeks. Custom VPS server based on HLoc (hloc + SuperPoint + SuperGlue + Colmap) without external dependencies: 3–5 months. Cost is calculated individually. Get a consultation — we will provide a preliminary estimate. Order VPS integration into your application.
We develop AR solutions for retail, museums, and industry. Contact us to discuss your task and select the optimal VPS solution.