nghorbani's picture
Update measurements/README.md
714498e verified
|
Raw
History Blame Contribute Delete
2.23 kB

Hyperspectral Data Log

Project: Distridays UseCase 1

Date: 2026-04-24

Operator: AI Team @ Cubert

Hardware

  • Camera: Ultris XMR
  • Lens: 50 mm / f2.0
  • Accessories: None

Setup & Lighting

  • Light source: Sunlight
  • Distance: ~100 m
  • Notes: Busy bus station. Human actors with similar outfits looking different in spectral view.

Acquisition Settings

  • Measurement mode: Video
  • Integration time: 17 ms
  • Frame rate: 15 fps
  • White reference: None
  • Dark current: Recorded

Scene Description

Object type: Human (complex scene)

Goal: Demonstrate that hyperspectral features beyond the visible spectrum improve person tracking in a crowded scene — fewer false ID switches and more reliable lock on the correct person than an RGB-only tracker can achieve.

Setting: Outdoor bus station with a busy pedestrian background. Camera on a static tripod roughly 100 m from the subject area, pointing at a section where pedestrians pass and wait. Hyperspectral data cube captured in video mode; spectral bands beyond the visible are exploited as tracking cues.

Subjects: Three actors — T (target), D1 (decoy 1), D2 (decoy 2). All three wear visually identical outfits so they look similar in the RGB projection of the hyperspectral cube. T wears a garment made of a material that looks distinct in CIR (colour infrared). This CIR difference is the spectral cue the tracker is expected to exploit.

Passive phase: Actors walk in single file toward or away from the camera, occluding one another along the optical axis. A visible-only tracker tends to swap IDs in this setting; hyperspectral cues should preserve the correct lock.

Active phase: While on camera, T carries a bottle of "invisible" ink that is visible only in specific spectral bands and sprays D1 with it. The spray adds a trackable spectral signature to D1 mid-recording. After marking, the three actors walk out of the scene from the opposite side, so the tail of the take shows the post-marking state.

Credits

Recorded and processed by the AI Team@Cubert. Use-case demo at http://docs.cuvis.ai/