Joint tracking and locomotion state recognition of C. elegans from time-lapse image sequences

There is a continued need for improved automated algorithms for tracking the movement of C. elegans worms from time-lapse image sequences, computing measurements, and identifying specific states of worm locomotion. The tracking and locomotion state recognition have been addressed sequentially in the prior literature. However, knowing the locomotion state can help predict worm dynamics while improved worm tracking can allow one to infer worm locomotion state more accurately. To exploit this obvious but unexploited synergy, this paper presents a 3-level model for simultaneous tracking and locomotion state recognition. Use of this model is shown to result in improved tracking performance compared to previously reported methods.