The Three-Layer Capture Method: Small Rooms to Looped Routes

A single-height orbit only reconstructs information near one horizontal slice, leaving top and bottom surfaces poorly covered. Shooting at three heights — high (tilted up), mid (eye-level), low (tilted down) — creates vertical overlap that dramatically reduces top or bottom dropout in the final reconstruction. Three layers is an engineering heuristic, not a hard rule.

SfM relies on recognizing the same regions across different frames to form 'loops'. Walking a wall-hugging closed loop — starting and ending near the same point — lets the tail of the sequence re-observe areas reconstructed early, enabling global bundle adjustment and reducing drift. Open-ended paths accumulate positional error toward the end, skewing the overall point cloud.

In small rooms (bedroom, hallway), the photographer and equipment easily enter the frame. Move backward along walls to keep your body out of frame; if using a selfie stick, control the shaft angle so its shadow doesn't appear on floors or walls. When self-occlusion is unavoidable, complete the three-ring route in the more open area first, then add supplemental shots in tight corners. Cloud training pipelines can often auto-mask dynamic objects, but prevention is more reliable.

Using a roughly 15 m² bedroom as an example: stand in the center, confirm three height levels and a starting point. High ring: from bed-height level, clockwise along walls (~40-50 frames). Mid ring: waist height, one full loop. Low ring: near floor level, one full loop. After three rings, return near the start and shoot 5-10 extra frames to reinforce loop closure. Scaniverse's live preview helps you judge reconstruction status; Luma gives alignment feedback after upload. Neither app enforces a specific path — adapt freely to your actual space. [unverified]