Implementation And Application Of Human Pose Estimation System Based On Computer Vision

With the improvement of software and hardware performance,artificial intelligence has developed rapidly in recent years.As a significant branch of artificial intelligence,computer vision has a wide range of commercial value and application prospects.Human pose estimation aims to estimate the position of key points of the human body from the image or video to form a complete human pose.Based on computer vision,human pose estimation covers a wide range of technical areas,such as sensor technology,machine vision and pattern recognition.It has good application value in human-computer interaction,sports analysis,rehabilitation,film and game production and other fields.However,the application of human pose estimation algorithms to social life is often restricted by the complexity in deep neural network for training and prediction.If we can centralize computing resources and provide services remotely,it will be more convenient for the application.This topic comes from the enterprise cooperation project,which contains designing and implementing a system for human pose estimation based on computer vision,and applying the system to rehabilitation filed.The main contents are as follows:We design and implement the system for human pose estimation.This system has the functions of image collection,transmission,algorithm analysis and visualization,which aims to provide 2D and 3D human pose estimation services in form of outputting human body keypoints.In this thesis,we design and implement this system in hierarchical structure,which contains data collection layer,data transport layer,algorithmic calculation layer,management layer,and visualization layer.Specifically,the data collection layer supports two methods for capturing images:grabbing images from web camera and uploading video files directly.After compressing and encoding,data transport layer will transfer data to computing server via message-oriented middleware,and perform algorithmic calculations in calculation layer,such as human pose estimation.Management layer helps to organize the algorithmic service for multiple users and multiple tasks.Finally,the results will be delivered to front-end in the form of image and body model in Unity in visualization layer.In this thesis,we apply the human pose estimation system to rehabilitation field and build rehabilitation training application,which contains the pose matching method based on human pose estimation and corresponding business functions,such as training plan,training reports and training evaluation.This application of the system helps to evaluate the movement of patients and provides assistance to trainers.This thesis demonstrates the process of architecture design and function realization.Additionally,it analyzes and verifies the function and performance through testing.