Occluded Gait Emotion Recognition Based on Interpolated Graph Convolutional Networks

In recent years, gait-based emotion recognition has been widely applied in the field of computer vision. However, existing methods for gait emotion recognition typically rely on complete human skeleton data, and their accuracy significantly declines when faced with occlusion. To improve the accuracy of gait emotion recognition under occlusion, this paper proposes an Interpolated Graph Convolutional Network(IGCN). Compared to the original graph convolutional network, the joint interpolation module employs a K-Nearest Neighbors (KNN) interpolation method to fill in the occluded skeleton joints in space, thereby reducing the negative impact of occlusion on emotion recognition outcomes. On the occluded dataset ELMB, the proposed method achieves an accuracy of 71.6%, significantly outperforming other methods, and alleviates the performance degradation caused by occlusion.