The present disclosure includes systems and methods for detecting and recognizing human gestures or activity in a group of 3D volumes using integral imaging. In exemplary embodiments, a camera array including multiple image capturing devices captures multiple images of a point-of-interest in a 3D scene. The multiple images can be used to reconstruct a group of 3D volumes of the point-of-interest using the images. The system detects spatiotemporal interest points (STIPs) in the group of 3D volumes and assigns descriptors to each of the plurality of STIPs. The descriptors are quantized into a number of visual words to create clusters of the group of 3D volumes. The system builds a histogram for each of the visual words. Each 3D volume is classified using the histogram of the plurality of visual words, implying the classification of a human gesture in each 3D volumes.