| Abstract [eng] |
This work deals with libraries or models based on artificial intelligence, which detect human body in images and video. These human pose estimation solutions open up new possibilities for the natural interaction between humans and computers, as these solutions can be used to create a system that responds to human gestures when visual material is received from a simple webcam. This paper thus additionally presents a computer game that is controlled by natural gestures, when a webcam is used to capture the image, and one of the libraries analysed is used to process the gestures. The first chapter reviews traditional methods that allow humans and computers to interact naturally. However, these methods require the purchase of special hardware. There are also alternative OpenPose, PoseNet, DensePose libraries (models) presented that do not require the purchase of special sensors. The second section describes the system developed that allows to control the created game with natural gestures. This section presents the system's use cases, requirements, deployment diagram, architecture and some game screenshots. The third chapter examines the PoseNet and OpenPose libraries. These studies are carried out using photos and videos. Furthermore, the results obtained are described and the advantages and disadvantages of PoseNet and OpenPose are presented. |
