Confidence Levels

EyesOnIt uses confidence levels to tell you how closely your object descriptions match your image or video frame. A confidence of 100 for an object description means that EyesOnIt believes the object description closely matches the image. A confidence of 0 means EyesOnIt sees little or no correlation between the object description and the image.

EyesOnIt always asks for at least two descriptions. It provides a confidence level for every object description. It does not provide confidence levels for background descriptions.

When we introduced object descriptions, we mentioned confidence levels. The image below shows a confidence level of 97% for the “vehicle” object description. No confidence level is shown for the background descriptions.

As you create your object descriptions, you want to help EyesOnIt achieve high confidence when the object that you describe is present, and low confidence when that object is not present. Let’s see how EyesOnIt uses these object descriptions to handle the case where no vehicle is present.

In the image below, you see that the confidence level of the “vehicle” object descriptions is 34% even when the vehicle is not present. This is much lower than the 97% confidence level that EyesOnIt reported when the vehicle was present. This large difference means EyesOnIt can accurately detect whether the vehicle is present or not present.

You may be surprised that EyesOnIt reports a confidence level of 34% even when the vehicle is not present. There are two primary reasons for this. The first reason is that the background description is being added to the object description. This was described in the section on object descriptions. EyesOnIt is internally using the description “vehicle driveway”. Since driveway is part of that description, the confidence of “vehicle driveway” is greater than zero. The second reason has to do with how computer vision models are trained. In the training data for EyesOnIt, many images of driveways also include vehicles. Because of this, there is some overlap between what EyesOnIt thinks a driveway looks like and what it thinks a vehicle looks like. But here’s the important point: the gap between the confidence levels when the vehicle is present and not present (97% - 34% = 63%) is high. When EyesOnIt generates a high gap in confidence levels between the cases where our object is present and not present, it can accurately detect the present or absence of the object.

As we will discuss in the next section, this accurate detection is critical for generating reliable alerts.