Simple evaluation of association quality in tracking-by-detection

Evaluating multiple-object trackers is challenging due to the variable number of quantities involved and the mixed discrete--continuum nature of the problem. Existing methodologies primarily address detection and tracking challenges. These challenges aim at the whole computer-vision pipelines as opposed to sole tracker algorithms. Modern tracker algorithms can get sophisticated enough to merit stand-alone analysis. The most critical part of the tracker is an association procedure. The outcome of the association procedure affects the tracking quality almost entirely. We propose a simple quality assessment framework to probe the association quality of trackers. The framework relies on a minimal, query-oriented instrumentation of the tracker. The instrumentation exposes the tracker internal association decisions allowing for a binary classification of the detection-target matches. The proposed methodology is easy to implement, requires little computational overhead and agrees with the well known ClearMOT metrics. The comparison is demonstrated through the open-source, ready-to-use software packages we provide.