Why SLS Cameras Give False Positives, and How to Stop Them

The biggest credibility problem in SLS ghost hunting is not skeptics. It is the camera itself drawing figures on a coat rack. If you want evidence people take seriously, you have to understand why this happens.

The model is guessing

Traditional SLS detection runs a skeletal tracking model built for gaming. That model is designed to find a body fast, even from incomplete data, because a game needs to react instantly. In a living room with a gamer standing in clear view, that works. In a dark, cluttered investigation site, it guesses, and it guesses wrong constantly.

A curtain with vertical folds reads as a torso. A chair back reads as shoulders. Your own arm passing through frame reads as a limb. The model has no way to know these are not people, so it draws a skeleton and moves on.

Why a confident-looking figure can still be noise

Here is the key insight. When the old system draws a figure, it does not tell you how internally consistent that detection is. A real human body produces a detection where most joints score similarly well: the network sees the whole figure clearly. A false detection produces what we call a Frankenstein read. A couple of joints score high while the rest score near zero. The system still draws a skeleton, but the geometry underneath is junk.

How SPECTER filters false figures

SPECTER scores every detected figure with a metric called Kinematic Coherence, on a scale from 0.0 to 1.0. It measures how consistent the joint confidence distribution is across a single figure.

• Above 0.7 means the figure's joints are internally consistent. It looks like a real body from a detection geometry standpoint, not noise assembling itself into a false positive.
• Low scores mean the detection is a Frankenstein read: a few strong joints, the rest empty. SPECTER scores it accordingly and will not escalate it to a detection event.

This is the difference between a system that draws a skeleton on anything and a system that tells you whether the skeleton is worth your attention.

Cross-reference before you log

Even with coherence scoring, good practice is to cross reference. SPECTER gives you two more SLS metrics for exactly this: Subjects Tracked, the count of distinct figures, and Average Confidence, the mean across all detected joints. A figure that should not exist in an empty room, with confidence sustained above 60 percent and coherence above 0.7, is a primary evidence indicator. One stray skeleton on a curtain is not.