The clustering is critical to understanding the multiscale behavior of fluidization. However, its time-resolved evolution on the particle level is seldom touched. Here, we explore both the time-averaged and time-resolved dynamics of clusters in a quasi-2D fluidized bed. Particle tracking velocimetry is adopted and then clusters are identified by using the Voronoi analysis. The time-averaged results show that the cluster hydrodynamic parameters depend highly on the cluster size and the distance from the wall. The number distribution of the cluster size follows a power law (~nc-2.2)) of the percolation theory except for large clusters (nc>100). The time-resolved analysis shows that the cluster coalescence can be simplified as a collision between two inelastic clusters, during which the net external force is roughly zero, and a snowplow model is proposed to predict its energy loss, ΔE ~ t3/2. The cluster rupture is suggested to be caused by increasing torque.