Recently, efforts to help athletes develop skills and abilities by incorporating scientific knowledge are actively deployed in the field of sports. At the same time, as judging and scoring for increasingly sophisticated skills in scoring competitions such as gymnastics is becoming more difficult every year, the need to incorporate technologies to improve accuracy and fairness in a judging and scoring system is increasing accordingly. Therefore, Fujitsu Laboratories is engaged in research and development aimed at establishing 3D sensing technology that accurately measures and digitizes complex human movements three-dimensionally using machine learning. 3D sensing technology is comprised of 3D laser sensor technology, which generates depth images based on the contours of the body surface to represent human movements, and skeleton recognition technology, which quickly extracts the 3D coordinates of joints from the depth images. Regarding 3D laser sensors, we describe a split projection/detection optical system with a micro electro mechanical systems (MEMS) mirror for laser emission and detection, which is capable of more than 10 times the number of pixels obtained by conventional light detection and ranging (LIDAR) technology. We then demonstrate the usefulness of view-angle control technology for sports. We also describe the skeleton recognition technology, noting the fast, high-accuracy 3D joint coordinate extraction method based on a combination of skeleton recognition through machine learning and subsequent fitting. This paper illustrates the 3D sensing technology developed by Fujitsu Laboratories and presents experimental results of skeleton recognition in actual athletes' movements.
© Copyright 2018 Fujitsu scientific & technical journal. Kaikki oikeudet pidätetään. / Kääntänyt https://www.DeepL.com/Translator