Active colloids segmentation and tracking

Active colloids constitute a novel class of materials which have drawn a lot of attention in recent years. They are composed of spherical metal particles converting chemical energy into motility, mimicking micro-organisms. Understanding their collective behavior is key to applications. In this context, we address the problem of segmenting and tracking colloids in long video sequences corrupted with severe illumination changes. We propose a very accurate method to recover the individual trajectory of each colloid. First, a region-adaptive level set method is used to segment individual colloids or small clusters. Combining with the circular Hough transform further refines the segmentation. Second, we recover simultaneously all the colloids' trajectories using a modified min-cost/max flow method on a weighted graph with colloids as vertices. No motion regularity is assumed to define graph edges and their cost. The proposed method is evaluated on a real benchmark composed of nine video sequences with annotations. In terms of CLEAR MOT metric – a standard metric for evaluating multiple target tracking algorithms – our approach outperforms very significantly four standard methods.