Pseudo 3D Pose Recognition Network

Multi-view human pose recognition has been extensively studied in computer vision due to its significant practical implications. Nonetheless, it remains a challenging task effectively integrate distinctive view-based features and perform thorough qualitative analysis quantitative evaluations. In this paper, based on an innovative multi-view fusion module novel Mutable Scaling Shortcut Connection, pseudo 3D neural network was meticulously crafted. The proposed framework comprises four modules: Front Residual Module, Convolution Cross View Fusion Rear Detection Module. Module serves as the head with incipient heatmaps extraction functionality, taking preprocessed images of various views separate inputs. performs convolution for output from each view, enabling benefit other consequently. extracts deeper-level features, ultimately classification recognition. can be trained end-to-end evaluated Self-Built Multi-View dataset. Analytical evaluation approaches were used explain contributory effects which significantly improve accuracy approximately 70% 91%-94% through Feature Aggregation, Strong Interaction Property among views, Sparsity Reduction, Increasing Euclidean Distance.