Why a mirror can look like an opening to a camera
Many vision systems estimate depth by looking for clues on surfaces. A wall, for example, may show texture that becomes smaller or more tightly packed as it recedes. From those patterns, software can infer whether a surface is near or far—even in a single image, without measuring distance directly.
A mirror disrupts that logic. Its glass may be physically flat, but the image in it contains the visual patterns of whatever it reflects: a hallway, furniture, or a distant window. The system can interpret those patterns as evidence of real space behind the glass, rather than as a picture appearing on a nearby surface.
The reflection supplies the wrong depth cues
To a person, a mirror is familiar enough that its reflection is usually easy to recognize. A machine-vision model may instead rely on learned visual regularities: textures and edges that often indicate depth in ordinary scenes. In a reflection, those cues still look plausible, but they describe the reflected room—not the mirror’s position.
This can produce a depth map, an image in which each pixel represents an estimated distance, that places the apparent hallway or room beyond the glass. The error is not that the camera sees a literal opening; it is that the software assigns a convincing depth interpretation to misleading visual evidence.
Why it matters in a real room
A robot or other system using that estimate to navigate could misjudge where a surface begins or how far away an apparent passage lies. Recognizing mirrors therefore requires more than detecting surface texture: a system needs ways to distinguish reflected scenes from the spaces those scenes depict. A flat sheet of glass can show a deep room without making that room any closer.
