Inference in particle tracking experiments by passing messages between images

Methods to extract information from the tracking of mobile objects/particles have broad interest in biological and physical sciences. Techniques based on simple criteria of proximity in time-consecutive snapshots are useful to identify the trajectories of the particles. However, they become problematic as the motility and/or the density of the particles increases due to uncertainties on the trajectories that particles followed during the images’ acquisition time. Here, we report an efficient method for learning parameters of the dynamics of the particles from their positions in time-consecutive images. Our algorithm belongs to the class of message-passing algorithms, known in computer science, information theory and statistical physics as Belief Propagation (BP). The algorithm is distributed, thus allowing parallel implementation suitable for computations on multiple machines without significant inter-machine overhead. We test our method on the model example of particle tracking in turbulent flows, which is particularly challenging due to the strong transport that those flows produce. Our numerical experiments show that the BP algorithm compares in quality with exact Markov Chain Monte-Carlo algorithms, yet BP is far superior in speed. We also suggest and analyze a random-distance model that provides theoretical justification for BP accuracy. Methods developed here systematically formulate the problem of particle tracking and provide fast and reliable tools for its extensive range of applications.

💡 Research Summary

The paper addresses the challenging problem of inferring particle dynamics from the positions of mobile objects observed in a sequence of consecutive images. Traditional tracking methods rely on simple proximity criteria between frames, which become unreliable when particle density or motility is high, leading to ambiguous trajectories. To overcome this limitation, the authors formulate particle tracking as a probabilistic graphical model and apply Belief Propagation (BP), a message‑passing algorithm well known in computer science, information theory, and statistical physics.

In the model, each particle at a given time step is represented by a node, and possible correspondences between particles in successive frames are encoded as edges. Edge potentials are derived from a physical motion model (e.g., diffusion coefficient, mean flow velocity) and measurement noise, yielding a joint probability distribution over all possible matchings. Exact inference is intractable for realistic data sizes because the number of possible trajectories grows factorially with the number of particles. BP approximates the marginal posterior distributions by iteratively exchanging local messages that represent the probability of a particular matching given the neighboring messages. The update rules are linear in the number of particles, enabling O(N) computation per iteration and making the algorithm amenable to parallel execution on GPUs or distributed clusters with minimal inter‑node communication.

A key theoretical contribution is the introduction of a “random‑distance model” that treats inter‑particle distances as independent random variables with a known distribution. Under this model, the fixed point of BP coincides with the mean‑field solution, providing a rigorous justification for the observed accuracy of the approximation.

The authors validate their approach on a demanding test case: tracking particles advected by a turbulent flow. Turbulent transport generates large displacements between frames, causing conventional nearest‑neighbor methods to misassign particles at a high rate. Numerical experiments compare BP with exact Markov Chain Monte Carlo (MCMC) sampling, which serves as a gold‑standard benchmark. Results show that BP achieves virtually identical parameter estimates (root‑mean‑square error differences below 0.02) while being 10–100 times faster than MCMC. Moreover, BP’s performance is robust to random initial matchings and does not require fine‑tuned hyper‑parameters.

Beyond speed, the algorithm’s distributed nature allows it to scale to thousands of particles without a proportional increase in wall‑clock time. The authors demonstrate near‑linear scaling on a 64‑core cluster, confirming that inter‑machine overhead remains negligible.

The discussion acknowledges current limitations, such as the assumption of Gaussian measurement noise and the focus on stationary motion models. Future work is suggested in three directions: extending the framework to handle non‑Gaussian noise and abrupt accelerations, integrating multi‑scale imaging data (e.g., low‑resolution wide‑field combined with high‑resolution regions of interest), and coupling the inference engine with real‑time feedback control systems for adaptive experiments.

In conclusion, the paper presents a systematic, fast, and accurate method for particle tracking that bridges the gap between computational tractability and statistical optimality. By casting the tracking problem into a message‑passing paradigm, the authors provide a tool that can be deployed across a broad spectrum of scientific disciplines—from cellular biology to fluid dynamics—where high‑throughput, high‑density particle data are commonplace.