Uma ferramenta para estimativa de movimentos a partir de vídeos

Abstract

The objective of this work is to perform, using computer vision, the estimation of human motion, which consists of the process of identifying the movements of people through data inputs such as videos, using the model of Convolutional Neural Networks (CNNs), which are computational models inspired by the human brain. For this, the OpenPose was chosen as the tool, given that its architecture is based on CNNs. It identifies joints and connections between key points of the body, generating representations in the form of 2D or 3D skeletons. OpenPose is an open-source tool that has an active community of developers and has been applied in human motion analysis, gesture recognition, and human-computer interaction. The utilized database consisted of videos collected from YouTube, allowing an analysis of head movements in different contexts. The proposed method combines OpenPose with a code layer to automate the animation of character bust. Processing the data using OpenPose enabled the creation of realistic automatic animations, demonstrating its effectiveness and robustness in detecting key points, even under challenging conditions. Although there are limitations in animating the entire body, OpenPose and the code layer have proven to be useful in digital animation, with potential for applications in various contexts.