希賽
古慶宸
Athletes are constantly looking for innovative and accurate ways to improve their performance. A crucial part of improving sports performance relies on quick and accurate analysis of previous training sessions. This thesis aims to propose and develop an intuitive and efficient system for athlete movement assessment and visualization in multiple training sessions. This thesis also highlights all the challenges and experiments required to develop this kind of system, including the results and decisions taken along the way. The system incorporates optimized processes for extracting 2D/3D skeleton information from MP4 videos. Includes various tools to provide insightful analysis, encompassing angle visualization, body trajectory tracking, and a manual annotation tool enabling users to draw in a 3D environment within any browser. Additionally, various video alignment methods were evaluated, including our low-consumption yet competitive approach, alongside introducing a metric for assessing alignment accuracy when ground truth data is available. A novel recording web application was developed to address limitations imposed by browsers and operating systems’ regular screen recording, which also resulted in reducing storage requirements by an exceptional 89%. In conclusion, this thesis offers a significant contribution to athlete movement assessment and visualization through the development of an intuitive and efficient system. The system’s optimized skeleton extraction, diverse analysis tools, advanced video alignment methods, and innovative recording approach collectively enhance performance analysis and decision-making in sports training. By providing a user-friendly interface, compatibility with various devices, and several powerful features, the system empowers athletes and coaches to visualize, analyze, and exchange valuable feedback on training sessions, ultimately leading to improved athletic performance.
MIS Lab