Research on Human Body Features Extraction based on Attention Mechanism

Human body features are widely used in existing downstream applications including medical health, behaviour predictions. The existing features extraction methods are concentrated on the deep learning optimization process, which ignore the important locations on human body features and cause the trained learning model can not precisely extract the fundamental body features. In this work, we initially utilize objective detection algorithm to identify the target area of human body and submit the detection results to the feature extraction module. Subsequently, attention mechanism to concentrate graph convolutional neural network on the body joints and motivate the model to generate precise extraction results. Finally, we utilize the MSCOCO data-set to evaluate our proposed model and compare with most used feature extraction algorithms. From our extensive simulation results, we can significantly conclude that our proposed attention mechanism can achieve the effectiveness and high accurateness of body features extraction through comparing with existing extraction methods in the identical simulation environments and data-sets.