Human Action Recognition by Decision-Making Level Fusion Based on Spatial-Temporal Features