Action Recognition by Joint Spatial-Temporal Motion Feature

Figure 2

First row shows original video sequence for testing data (Lena_run1 from Weizmann dataset), and second row is background subtracted video sequence to calculate optical flow for training data.