Monotonicity and Error Type Differentiability in Performance Measures for Target Detection and Tracking in Video

There exists an abundance of systems and algorithms for multiple target detection and tracking in video, and many measures for evaluating the quality of their output have been proposed. The contribution of this paper lies in the following: first, it argues that such performance measures should have two fundamental properties – monotonicity and error type differentiability; second, it shows that the recently proposed measures do not have either of these properties and are thus less usable; third, it composes a set of simple measures, partly built on common practice, that does have these properties. The informativeness of the proposed set of performance measures is demonstrated through their application on face detection and tracking results.