This paper introduces *Motion History Volumes* (MHVs) as a free-viewpoint representation for human action recognition in multiple calibrated and background-subtracted video cameras. The authors present algorithms for computing, aligning, and comparing MHVs of different actions performed by different people from various viewpoints. Alignment and comparison are efficiently performed using Fourier transforms in cylindrical coordinates around the vertical axis. The results indicate that this representation can be used to learn and recognize basic human action classes, independent of gender, body size, and viewpoint. The paper is organized into several sections, including an introduction, definitions, motion history images, motion history volumes, temporal segmentation, motion descriptors, classification using motion descriptors, and conclusions. Key contributions include the extension of 2D motion templates to 3D, the use of Fourier-based features for invariance to rotation and translation, and the effectiveness of the proposed method in recognizing simple action classes from multiple viewpoints. The classification results using the IXMAS dataset demonstrate the method's ability to handle large datasets and realistic video sequences, achieving high recognition rates even with false positives.This paper introduces *Motion History Volumes* (MHVs) as a free-viewpoint representation for human action recognition in multiple calibrated and background-subtracted video cameras. The authors present algorithms for computing, aligning, and comparing MHVs of different actions performed by different people from various viewpoints. Alignment and comparison are efficiently performed using Fourier transforms in cylindrical coordinates around the vertical axis. The results indicate that this representation can be used to learn and recognize basic human action classes, independent of gender, body size, and viewpoint. The paper is organized into several sections, including an introduction, definitions, motion history images, motion history volumes, temporal segmentation, motion descriptors, classification using motion descriptors, and conclusions. Key contributions include the extension of 2D motion templates to 3D, the use of Fourier-based features for invariance to rotation and translation, and the effectiveness of the proposed method in recognizing simple action classes from multiple viewpoints. The classification results using the IXMAS dataset demonstrate the method's ability to handle large datasets and realistic video sequences, achieving high recognition rates even with false positives.