This paper presents a novel framework for recognizing and linking visually ambiguous events, specifically focusing on the problem of associating people with bicycles in a rack over multiple days. The authors propose a Bayesian network to model the posterior dependencies between atomic events (people and bicycle clusters) and compound events (drop and pick actions). The framework uses Reversible Jump Markov Chain Monte Carlo (RJMC) to sample feasible explanations, with simulated annealing to find the Maximum A Posteriori (MAP) solution. The method is evaluated on a challenging dataset, demonstrating improved accuracy compared to separate event recognition and linkage tasks. The framework is designed to handle multiple layers of event linkage and can be extended to other domains with related constraints.This paper presents a novel framework for recognizing and linking visually ambiguous events, specifically focusing on the problem of associating people with bicycles in a rack over multiple days. The authors propose a Bayesian network to model the posterior dependencies between atomic events (people and bicycle clusters) and compound events (drop and pick actions). The framework uses Reversible Jump Markov Chain Monte Carlo (RJMC) to sample feasible explanations, with simulated annealing to find the Maximum A Posteriori (MAP) solution. The method is evaluated on a challenging dataset, demonstrating improved accuracy compared to separate event recognition and linkage tasks. The framework is designed to handle multiple layers of event linkage and can be extended to other domains with related constraints.