Face detection

Purpose

Demonstrate enabling face mode on the online recognition pipeline, consuming quality-ranked FaceModels, and extracting optimised crops for downstream usage.

Entry points

Glass home:

• HomeActivity → GlassFaceTrackActivity

Sample Activity:

• com.rokid.glass.GlassFaceTrackActivity

Platform

• Glass

Main file

• GlassFaceTrackActivity.kt

Flow

1. Grab GlassSdk.getGlassOnlineRecService().
2. Attach IGlassDetectionListener.
3. startDetection(MODE_FACE).
4. Consume onProcessedFaceModels() to pick the best samples.
5. Filter noisy tracks via quality + trackId.
6. Snapshot crops through getFaceSamllBitmap(trackId) and render.

FAQ

What “best face” means here

Heuristic blend of bounding-box area, iqaScore, faceScore, deduped per trackId; samples under iqaScore < 40 drop out.

Detection vs recognition

This path emphasises detection + crop quality scoring—no ID compare is performed in UI.

Caveats

• Always call stop + detach on exit.
• Persistence hooks exist but stay commented—focus is pipeline validation, not archival UX.