Skeleton-Based Action Recognition Based on Distance Vector and Multihigh View Adaptive Networks

Figure 2

The best keypoint representation under the same action at different height angles.