Deep dive
User guidance layer.
The guidance layer closes the loop by telling the user how to improve the camera view when the system cannot read confidently.
Purpose
For visually impaired users, a camera interface cannot assume the user can inspect the preview. Guidance converts visual quality problems into spoken instructions.
- Explain why reading failed instead of only saying it failed.
- Give short corrective prompts during capture.
- Use quality checks to decide when guidance is needed.
- Help the user build a better frame without visual feedback.
Implementation direction
Guidance can be generated from measurable signals such as blur, brightness, glare, crop confidence, and OCR confidence. The system can map each problem to a short spoken prompt.
- Low blur score can trigger “hold steady.”
- Small text size can trigger “move closer.”
- Glare detection can trigger “tilt the page away from the light.”
- Low page confidence can trigger “center the page in view.”
Failure modes
Guidance becomes annoying if it is too frequent, too vague, or wrong. The system needs to avoid repeating the same message continuously and should prioritize the most important issue.
- Repeating prompts can frustrate the user.
- Too many instructions at once can be hard to act on.
- Incorrect guidance can make capture worse.
- Late guidance slows down the reading experience.
Future improvements
A polished version would use short, prioritized prompts and only speak when the user can act on the message. Guidance should feel like a calm assistant, not a warning system.
- Add cooldowns so prompts do not repeat constantly.
- Prioritize one fix at a time.
- Use audio tones for simple alignment feedback.
- Learn from successful captures to improve future prompts.