Environmental effects should be tested with the target scene's real lighting before deployment, not laboratory ideal lighting

What it is

Uniform, windowless, stable-CCT lamps in a lab will measure gesture recognition as a flattering accuracy. That number does not replace a lighting test in the target scene. Before deployment, run the same tasks in the kind of room that will be installed, at the hours and weather it will actually see. Ideal lighting is a calibration condition, not a product promise. That environment affects recognition is already established; what is required here is that evidence come from the site, not from a fixture catalogue.

Why it happens

Labs, for repeatability, remove windows, glossy tables, mixed sources, and walking people, and thereby pin the reliability function at a convenient point. A real venue walks that point through the day and stacks occlusion and clothing on it. Vendor evaluation video is almost always shot at that convenient point. If acceptance sits on the same point, first-week field failures get billed as “users don’t know it,” which is distribution shift. Field testing need not carry every lab independent variable; it needs combinations the site will meet: morning backlight, noon sun, evening ceiling lights, opening-hour traffic. Miss the combination, miss the product.

Studying it

On the same tokens and the same device, measure under lab standard lamps and at several hours in the target venue, and report the gaps in accuracy, keypoint error, and failure type. Take a “passed in lab only” build to the field and watch a week of tickets. Methodologically this is a transfer evaluation, not another lux sweep. Log weather and whether blinds were open as part of the field condition, not as noise.

Where it stops holding

A windowless control room whose fixtures match the lab can sit close to the lab, and should still be measured once after the final lamps are in, because real spectra drift. A touring exhibit changes venue daily; one field test is not enough—define an acceptable venue lighting envelope and spot-check each stop. A solar simulator can add backlight to an indoor lab and still lacks a real window’s geometry and walk through the day. User studies recruited only in a university lab will also systematically miss field light.

Applying it

• Put “target venue × at least two extreme usual hours” of task video on the release checklist. Lab numbers are a baseline only.
• When contracting hardware metrics, write the field conditions. Refuse dark-room evaluation video as the only evidence.
• In week one of launch, book failures by hour and weather. If failures follow the sun, change install and the test plan before blaming the vocabulary.

Related

• Same group: C4.31.1 Reliability of vision-based gesture recognition varies with ambient light intensity and direction · C4.31.2 Strong backlight or overly dark environments significantly reduce skeleton and hand-keypoint accuracy · C4.31.3 Occlusion by obstacles or other bodies interrupts continuous motion tracking
• Adjacent: C4.25 Predefined fallback paths · C4.05 Three dimensions of gesture evaluation
• Search: in situ · field lighting · lab-to-field