Quantifying the heterogeneous gas-solid flow characteristics is of great significance to the design and optimization of this type of reactors. The formation and evolution of particle clusters result in the breakdown of the local equilibrium hypothesis, leading to the non-Maxwellian particle velocity distribution and the fail of classical kinetic theory of granular flow (KTGF). The particle movement characteristics and velocity distribution function in a fast fluidized bed were analyzed by using the experimental and data processing methods combing adaptive particle image velocimetry (PIV), particle tracking velocimetry (PTV) and digital image processing (DIA). It was found that the particle velocity distribution obviously deviated from the Maxwell distribution in both the gravity direction and the horizontal direction, showing a long-tailed off-peak or bimodal distribution. The bimodal distribution function can be used to fit the particle velocity distribution pretty well. In this work, the influence factors of anisotropy and non-Gaussian distribution characteristics of particle velocity distribution were further discussed. The particle velocity distribution function showed significant anisotropy in the horizontal and vertical directions. The deviation of particle velocity distribution function from the ideal distribution was positively correlated with the degree of particle clustering, the local particle velocity fluctuation and the particle concentration fluctuation, which may exhibit bimodal distributions in the near-wall regions. In order to analyze the relationship between particle velocity distribution function and mesoscale structure of fluidized bed more comprehensively, the energy-minimization multi-scale (EMMS) model was adopted to calculate the heterogeneous parameters. The consistency between the experimental results and the hypothesis of the dilute-dense phase coexistence showed the theoretical feasibility of the bimodal distribution.