Multi-Scale Locality-Constrained Spatiotemporal Coding for Local Feature Based Human Action Recognition

Figure 2

The illustration of our method. The shape of code words indicates appearance information, while the color indicates spatiotemporal position. For example, the code words ■□ have same appearance but different position. They are considered as two different code words in feature coding. Then, two different code words histograms are obtained for two actions upon these code words. Hence, Actions 1 and 2 which cannot be distinguished in BoF (Figure 1) can be correctly classified with our method.