Reading AYNIcam detection boxes and confidence

Start with the camera image

Detection keeps the camera image visible and adds boxes around supported objects. Before driving, mount the iPhone securely, clear the lens and check that the road is in view. Select a detector model and open Detection. The image remains your reference when a box is late, incomplete or misplaced; the overlay is an interpretation of that image.

Know what the model can label

The detector supports person, bicycle, motorcycle, car, bus, traffic light and stop sign. Both nawnaw and nawnaw pro use this seven-class set. A vehicle outside these labels can be missed or assigned an imperfect label. A box around a car does not identify its make, prove which end is the front or measure its exact size.

Read confidence in context

Confidence describes the model’s support for a detection. It is not the probability of a collision, a certified measurement or a promise that every higher score is correct. A partly hidden bus may receive a changing score while a reflection may still produce a false detection. Evaluate the visible object and the continuity of observations together.

When boxes change or disappear

Occlusion, darkness, glare, small distant objects and camera movement can change the result. A box disappearing does not mean the road is clear. Traffic-light recognition can also be uncertain: always follow the actual signal. If results become distracting, change your view or settings while safely parked. Never adjust a detection threshold or inspect small confidence values while driving.

Choose the view for the task

Use Detection to compare an overlay with the source image. Virtual View simplifies the same observations; it does not create a second independent confirmation. Segmentation adds object masks and a road-surface output with its dedicated Pro model. None of these views supplies a complete account of the road outside the camera’s field of view.