Schematic examples of different tracking scenarios. For simplicity, only the elements of the overlapping sets are shown for each cluster. A Clusters \(\tilde C^{t_2}_1\) and \(\tilde C^{t_2}_2\) generate a match with classes \(C^{t_1}_1\)and \(C^{t_1}_2\)as the overlapping sets lead to large overlapping values. B In this case, clusters \(\tilde C^{t_2}_2\) and \(\tilde C^{t_2}_3\) are matches with classes \(C^{t_1}_2\)and \(C_1^{t_1}\), respectively, but cluster \(\tilde C^{t_2}_1\) is labeled as a split with respect to class \(C_1^{t_1}\). Since clusters \(\tilde C^{t_2}_3\) and \(\tilde C^{t_2}_1\) together include the majority of the spikes of \(OS(C_1^{t_1})\), they are automatically combined and included in class \(C_1^{t_1}\) to generate class \(C_1^{t_2}\). C For this example, \(\tilde C^{t_2}_2\) is a match with class \(C^{t_1}_2\), but clusters \(\tilde C^{t_2}_1\) and \(\tilde C^{t_2}_3\) are labeled as new classes. Cluster \(\tilde C^{t_2}_1\) is a split of class \(C^{t_1}_1\), but without another split or match it cannot be included in the class. Cluster \(\tilde C^{t_2}_3\) has a reasonable overlap with both classes (so it cannot be a split) but not large enough to generate matches. D In the left panel, cluster \(\tilde C^{t_2}_1\) is labeled as a temporary merger at block t2 (associated to classes \(C^{t_1}_1\)and \(C^{t_1}_2\), that define its ML list). In the next block, the same overlapping sets will be used (i.e. \(OS(C_1^{t_2})\ =\ OS(C_1^{t_1})\) and \(OS(C_2^{t_2})\ =\ OS(C_2^{t_1})\)). The middle panel shows that during block t3, clusters \(\tilde C^{t_3}_1\) and \(\tilde C^{t_3}_2\) were matched to the previous classes, so the merger is not confirmed and cluster \(\tilde C^{t_2}_1\) will be later resolved as explained below in the Graph Reduction phase (see Fig. \ref{748245}D). Alternatively, as shown in the right panel, cluster \(\tilde C^{t_3}_1\) is another merger associated to the same ML list. If this situation is repeated for \(MS\) consecutive blocks, the merger is confirmed, a new class is associated to the merged cluster in all \(MS\) blocks, and the individual classes from the ML list "disappear".