Features are an important component for the majority of machine learning based techniques. Furthermore, voxel based features are features that convey important meaning from 3D space. This project considered multiple approaches to how voxel based features can be estimated. This included from point clouds derived from e.g. depth based cameras and also more directly via 3D imaging modalities such as X-ray Computed Tomography (XCT). A number of unique 2D and 3D imaging features have been developed. This includes an information theoretic 3D feature descriptor that describes the irregularity of the orientation distribution of fibres in a range of materials. Other new voxel based features to be derived are descriptors based on Motion History Images (MHI)s and Depth Motion Maps (DMM)s. These are 2D representations of 3D and 4D information respectively. They are enhanced sources of features, where additional multi-dimensional information is captured into appropriate reduced dimensional spaces. Deep learning has also been leveraged to take advantage of automated methods of feature learning and extraction.