Multi-Scale Locality-Constrained Spatiotemporal Coding for Local Feature Based Human Action Recognition
Figure 1
The illustration of that BoF ignores the spatiotemporal relationship of local features in STV. There are two groups of similar features with different spatiotemporal arrangements in STV. They are form Action 1 and Action 2, respectively. However, they cannot be correctly classified with BoF, because it obtains two same code words histograms due to ignoring feature spatiotemporal relationship.