A combination of the level set method and circular hough transform to segment highly cluttered colloids.
All the colloids' trajectories are recovered simultaneously within a single graph.
We annotate in total nine video sequences by human observers.
The proposed method reaches average accuracy of 93.19