This article proposes a human posture classification system to classify the four postures of standing, bending, sitting, and lying using three-dimensional (3-D) fuzzy body voxel features obtained from two cameras and hierarchical fuzzy classifiers (FCs). The system consists of four parts: human body segmentation, 3-D body voxel model construction, feature extraction, and classification. A region-dependent Gaussian mixture model with shadow removal is proposed for human body segmentation. The silhouette volume intersection method with computation speedup is then used to construct the 3-D body model using the segmented images from two cameras. After that, fuzzy 3-D body voxel features for posture classification are extracted from a minimum enclosed cube of the 3-D model. For classification, this article proposes an FC with soft margin minimization (FC-SMM) and its hierarchical structure to improve the classification performance. The consequent and antecedent parameters are optimized through a linear batch support vector machine and margin selective gradient descent algorithm, respectively. The FC-SMM also shows the advantage of model interpretability through visualization of the fuzzy rules. Experimental results of the comparison of different feature extraction methods and classifiers in different videos show the superiority of the proposed classification method.