Skeleton-Based Adaptive Graph Convolutional Networks for Cockpit Sign Language Classification

To solve the problem of barrier-free communication for deaf-mute people in traffic travel, the problem of cockpit sign language classification is proposed. On this issue, this paper proposes an adaptive graph convolutional network based on skeleton features (NA-AGCN), which is a way to improve the ST-GCN model. First of all, a lightweight normalized attention module NAM is added, which allows the network to suppress those insignificant weights, thereby improving network efficiency; then, an adaptive graph convolution module is proposed, which breaks the original graph The constraints of the convolutional adjacency matrix divide the matrix into three subgraphs, realize the extraction of global features, and increase the flexibility of the model construction graph. The model has achieved a good classification effect on the self-made cockpit sign language data set, and can accurately obtain the sign language expression results of the people in the video. This model has very good practical significance and can enable deaf-mute people to better enter society. Integrate into society and realize the effect of barrier-free communication between deaf and mute people in transportation and travel.