Many works of human posture recognition have been published in the literature. Gesture and posture recognition by using Kinect Sensor is an important technique for many types of application. In this paper, we apply a neuro-fuzzy system for classification of human posture. A recent approach of learning algorithm for Single-layer hidden feedforward neural network (SLFN), known as Randomized Neural Network (RNN), has been attracting attention as a solution of several problems of traditional neural networks. However, the performance of RNN varies depending on the number of hidden neurons. In this paper, ensemble learning methods are therefore developed to deal with the issue, using Fuzzy-RNNs with various sizes of networks. Moreover, we utilize evolutionary algorithm to optimize the structure of ensembles. This paper discusses the effectiveness of the proposed approach, comparing with performance of some ensemble learning methods by using benchmark datasets and real measurement data of human posture.