Two measures of performance are relevant to Pfinder. The first is classification performance. This is a measure of how well Pfinder is classifying the scene, and will be examined below by measuring the error and noise in the estimates of body-part positions that are based on the classification output.
Since Pfinder is intended to be an interactive system, the other important measure is the runtime performance. This is measured below in terms of per-frame and per-pixel computation rates.