In this paper, a fused random forest algorithm named (PSO-RF)-(KNN-HC) is proposed for the recognition of seven human motion patterns, including flat walking, sitting, standing, going up the stairs, going down the stairs, going up the slope and going down the slope. A particle swarm optimization (PSO) method is used to find the optimal parameters of the random forest model and build the optimal classification model. In the decision process of the random forest, the algorithm of k-nearest neighbors-hierarchical clustering (KNN-HC) is applied to select the decision trees for new recognition samples and calculate the voting weights of each tree, which improves the classification accuracy of the random forest model for multi-classification problems. In the data processing stage, the motion data are analyzed from view of the frequency domain using the fast Fourier transform (FFT) to divide the data segments in cycles and perform feature extraction. Finally, the proposed algorithm is validated against other machine learning algorithms based on a self-constructed human motion dataset through a real motion data acquisition platform, and the effectiveness of the proposed method is also validated on an open source dataset.