Fig. 5From: Action recognition using length-variable edge trajectory and spatio-temporal motion skeleton descriptorVisualization of proposed LV-ET. From top to bottom: brush hair, riding horse, and diving. The above two rows correspond the case of relatively static action with slight camera motion. The bottom row corresponds to the drastic camera motion case, lots of points moving together in one frame. From left to right: a frame of an action video, the trajectories unpruned and the trajectories prunedBack to article page