Deep Neural Networkを応用した画像認識によるスポーツ用動作解析システムの基礎的研究

抄録

Recently, motion analysis systems have been used in various fields such as medical field, entertainment, and sports. There are some types of motion analysis systems that use cameras and force plates, and use inertial sensors, etc., but the common problems are the experiments in the restricted environment and errors caused by muscular contraction. Therefore, it is difficult to analyze the motion of athletes who are in the actual games. Therefore, we propose a novel motion analysis system using a solid model. On this method, 3D motion analysis data can be acquired from a two-dimensional image by using a 3D solid model. As a result, this method allows motion analysis based on general broadcast video, thus motion analysis of athletes in the actual games can be performed. However, the operation time is enormous, because the matching manual work required for converting the two-dimensional image into the three-dimensional motion analysis data is manually performed for each frame of the video image. Therefore, the objective of this research is to automate the matching work to shorten the time for matching work. In this research, we develop a system that automates matching work. As a matching work of alpine ski turn movement, we try to automate matching work by using a discriminator by DNN which is drawing attention in the field of image recognition.