A Hybrid Multi-Perspective Complementary Model for Human Skeleton-Based Action Recognition

Owing to the efficiency of skeleton data, tasks involving skeleton-based action recognition have been receiving in-creasing attention. Among them, achieving accurate action recognition and classification of human skeleton data from the perspective of unmanned aerial vehicles (UAVs) is one of the crucial tasks. However, most prior skeleton-based action recognition methods tend to focus on a singular mode of skeleton information transmission and interaction, overlooking the potential complementarity amidst di-verse skeleton data and various modeling techniques. To address this, we present the Multi-perspective Complementary Model for Human Skeleton-based Action Recognition. This model recognizes and interprets human skeleton data from various angles, extensively leveraging the complementary fusion of different views and models to enhance action recognition performance and conducts extensive experiments on the UAV-Human dataset. The results show the su-perior effectiveness of the proposed method, achieving top 3 performance on the leaderboard. The code is available at https://github. com/happylinze/UAV-SAR.