A Comparison of Fusion Techniques for Multi-Modal Human Activity Recognition on the HARMES Dataset