Fig. 4From: Multi-surface analysis for human action recognition in videoFramework of the HOG feature. The MHI is divided to M × N grids. The gradient of pixel casts a weight vote for an orientation-based histogramBack to article page