End-to-end hand object interaction attitude estimation method and system

The invention belongs to the technical field of three-dimensional attitude estimation, and discloses an end-to-end hand object interaction attitude estimation method and system, the method comprises a feature extraction stage and a reconstruction stage, the feature extraction stage uses a key point detection model to extract features and a heat map; the reconstruction stage comprises a posture reconstruction branch and a shape reconstruction branch, the posture reconstruction branch reconstructs the posture of the hand and the object in the interaction scene, and the shape reconstruction branch reconstructs the shape of the hand and the object in the interaction scene. According to the method, an improved GraFormer module is adopted to improve a posture reconstruction branch and a shape reconstruction branch, a semantic graph convolution layer is added into the improved GraFormer module, semantic information between adjacent nodes is extracted through the semantic graph convolution layer, and the semantic exp