Research And Implementation Of Human Pose Estimation Technology Based On Deep Learning

The field of artificial intelligence continues to advance under the development of deep learning,and nowadays deep learning algorithms have been widely used in all walks of life,from product recommendation to face recognition,the role of artificial intelligence can be seen everywhere.Computer vision is one of the important fields of artificial intelligence.Traditional image processing methods for human pose estimation often use manual extraction of features and fusion of the extracted features,then analyze the relevant human key points and finally use the graph structure to match the human pose.Human pose estimation has been greatly promoted by the application of deep learning algorithms,and there exists a very rich application area.In this paper,we try to develop a software library that can cover the whole development process of the human pose estimation problem,providing corresponding components for each aspect of data acquisition,sample generation,data augmentation,loss function,model training and model inference,and providing a variety of backbone models and pose estimation models to make the whole development process of human pose estimation software more convenient.This software attempts to develop a tool library for human pose estimation based on PyTorch as the base deep learning framework.The goal of the software is to satisfy every need of developers in the human pose estimation software development process and provide a series of clear and friendly program interfaces,and the software will be open-sourced to facilitate developers to make more in-depth modifications and secondary development.This work validates the feasibility of the Py Torch-based human pose estimation tool library and provides a basis for the development of other Py Torch-based computer vision software tools,and gives some reference value to related tools.