How We Built a High‑Performance Geospatial Photo Index for a Travel App

Part 1: Application Scenario and Requirements

To let users quickly find photos or videos they want to share, the service must aggregate and sort albums by destination and creation time, and support radius‑based geographic searches.

Part 2: Design and Algorithm Selection

Initially a brute‑force server‑side traversal was considered, but the large data volume and duplicate clusters made it impractical. After evaluating common spatial indexing methods, GeoHash and Google‑S2 were compared. GeoHash suffers from abrupt ordering changes that reduce precision, while S2 offers smooth, high‑resolution cells (30 levels) and better accuracy. The team therefore adopted the S2 algorithm for spatial point indexing.

Part 3: Implementation and Performance Optimizations

The service consists of four modules:

Album Index Data Operations : stores photo metadata and S2 Cell IDs (Level 4‑16) in a database.

User Album Scanning : uses iOS photo‑library APIs to detect added or removed photos.

Album Index Service : provides APIs for querying photos by Cell ID, level, or radius, and synchronizes with the scanning module.

Location Classification Module : maps Cell IDs to destination IDs (mdd_id) using server‑side aggregation and caches results.

The overall workflow updates the local index at app launch, serves cached classification results instantly, and refreshes them in the background.

Performance was further improved by:

Fetching incremental photos using the iOS "Recent" album, processing only the newest 100 items until an already‑indexed photo is found.

Progressively classifying photos: processing batches of 1,000 photos, assigning them to destinations on‑the‑fly, and rendering results incrementally to reduce cold‑start latency.

Part 4: Future Plans and Summary

The S2‑based geospatial album is already live in the iOS client, boosting note‑publishing volume. Ongoing work will refine database handling and explore deeper S2 capabilities, aiming to provide sub‑meter accuracy for more complex travel‑recording features.