A Continuous, Full-scope, Spatio-temporal Tracking Metric based on KL-divergence
This provides a continuous and comprehensive evaluation tool for researchers in computer vision and tracking, though it is incremental as it builds on existing metric concepts.
The paper tackles the problem of evaluating object tracking systems by proposing a unified metric based on KL-divergence, which handles errors like false alarms and splits, and shows results on the Oxford Town Centre Dataset with comparisons to existing metrics like MOTA.
A unified metric is given for the evaluation of object tracking systems. The metric is inspired by KL-divergence or relative entropy, which is commonly used to evaluate clustering techniques. Since tracking problems are fundamentally different from clustering, the components of KL-divergence are recast to handle various types of tracking errors (i.e., false alarms, missed detections, merges, splits). Scoring results are given on a standard tracking dataset (Oxford Town Centre Dataset), as well as several simulated scenarios. Also, this new metric is compared with several other metrics including the commonly used Multiple Object Tracking Accuracy metric. In the final section, advantages of this metric are given including the fact that it is continuous, parameter-less and comprehensive.