A camera that says what matters.
Point it at a room. Two YOLO models find obstacles, work out which way they are and how close, and the same decision rules as the Android app pick the one sentence worth saying aloud. Everything runs in this browser tab. No photo leaves your device.
Drop a photo here
or choose one of the examples below
Loading models…
BlindAssist says
—
Detections
Position
Each box becomes a clock bearing (12 o'clock is straight ahead) and a proximity bucket. Thresholds are per class, so a "close" bottle is a far smaller box than a "close" person.
Decision
Walk mode only speaks about obstacles that are close or very close, and says which side is freer. Low-confidence names are spoken as "obstacle" rather than risk a wrong noun.
Two models, one answer
A COCO YOLOv8n plus a fine-tune for doors, which COCO lacks. Overlapping boxes from the two are merged by IoU, so one object is never announced under two names.
A still-image demo of the Android app's pipeline; the app adds persistence and cooldown across frames, offline voice commands, sonar and OCR. Distances are rough estimates from box height. Not a safety device.