Best Camera Setup for Bite Detection

Setup beats hardware

People assume detection quality is mostly about camera quality. It is not. A standard laptop webcam in good conditions substantially outperforms an expensive camera in bad ones.

The models being used need to resolve landmarks on your hand and face. What determines whether they can is lighting, framing, and angle. Resolution beyond a certain point contributes very little — detection samples at a modest resolution by design, because that is ample for landmark placement and dramatically cheaper to process.

Lighting is the single biggest factor

If you fix one thing, fix this. The most common cause of missed detections is a face lit from behind.

Sitting with a window behind you puts your face in silhouette from the camera's perspective. Your eyes adapt and it looks fine to you; the camera sees a dark shape against a bright background and landmark detection degrades badly.

The fix is to have your main light source in front of you or to the side. A window you face works well. A lamp behind the screen works well. Overhead lighting alone is mediocre because it casts downward shadows across the face, which is exactly where the mouth landmarks are. If you work in the evening, a small light near the monitor makes a disproportionate difference.

Framing and angle

The goal is that both your face and the space your hands move through are inside the frame.

• Position the camera so your head sits in the upper-middle of the frame with some room below your chin. Hands approach from below, and a tightly cropped head shot cuts off the approach entirely.
• Camera at roughly eye level is ideal. A laptop on a desk pointing steeply upward is the most common suboptimal setup — it foreshortens the face and puts the mouth at an awkward angle.
• Sit at a normal working distance, around an arm's length. Very close and your hands are out of frame before they reach your mouth; very far and landmark precision drops.
• Check the preview and deliberately raise a hand to your mouth. If you cannot see the hand entering frame, the detector cannot either.

Common situations and what to do about them

A few setups need specific handling.

Dual monitors are a frequent problem: if the camera is on one screen and you spend most of your time looking at the other, you are presenting a profile rather than a face for hours. Put the camera on the screen you actually use, or centre it between them.

Standing desks change your height relative to a fixed camera, so a setup tuned while seated may frame your chest while standing. Adjust the tilt when you change position.

Laptop-only users who work with the screen tilted well back should be aware that this points the camera at the ceiling. A small correction to screen angle often resolves a lot of missed detections at once.

And if you wear a hat with a brim, expect shadowing across the face that meaningfully degrades landmark quality.

Verifying it works before relying on it

Worth spending two minutes on this rather than discovering a problem after a week of quiet logs.

Deliberately bring a fingertip to your mouth and hold it there for about a second. The alert should fire — the system requires sustained proximity across consecutive frames rather than a single instant, so a quick pass-by intentionally will not trigger it.

Test in the position you actually work in, not the position you adopt while testing. Test in your evening lighting as well as your daytime lighting, since those are frequently very different and evening is a high-frequency biting window for many people.

If it does not fire reliably, work through lighting first, then framing, then camera angle, then sensitivity. Adjusting sensitivity to compensate for a bad setup is treating the symptom — a properly lit, well-framed camera at the default setting will outperform a badly positioned one at maximum sensitivity.