First Cycle Evaluation
Second Cycle Evaluation
The evaluation results presented here are those produced after the termination of the ETISEO project. In fact, one of the main objectives of the project is to gain insights into the video surveillance algorithms as much as possible. Therefore, we, at INRIA, are continuing to work in this direction.
The groundtruth data contains not only mobile objects, but also some static objects (Note: we are referring to a special class of static objects, i.e., those that were moving previously and come to a stop). In the first- and second-round evalaution, the static objects have been considered in the performance evaluation. However, except for a few partners who made special efforts to detect static objects explicitly, as it is requested, the other partners ignored them by considering them as part of the background. Thereafter it is useful to (re-)evaluate the algorithms in this context.
Also, our parallel research work proposes some improvements in metrics definition and the ways different metrics are combined to produce the final results. For details, please refer to the paper published on AVSS'07.
The figures below present the latest evaluation results according to the paper. The original data can be found in the MS Excel file. The static objects, from both the groundtruth and the results of the partner, are removed. The performance index is a weighted sum of some important metrics for each task and the formular can be found in the above Excel file.