Human Posture Identification for Sport Training

  • 陳 冠宏

Student thesis: Master's Thesis


The modern people often ignore the importance of establishing an exercise routine With possibly incorrect actions during exercise not only the fitness goal is not reached but also people may be injured In view of this an accompanying trainer who provides real-time and appropriate guidance is necessary On the other hand there are already various approaches in place that support sport training In this work we utilize a markerless device for motion capture and then conduct subsequent human motion analysis Note that several previous studies focus on the comparison of a single posture to evaluate the correctness of a trainee’s movements However a workout program is usually a motion sequence containing different postures A single posture is not enough to be representative We thus propose to utilize both the LCS (longest common subsequences) and the DTW (dynamic time warping) algorithms for matching whole sequences A prototype system is also implemented in which a user can imitate the postures as demonstrated by the trainer Specifically our prototype system provides functionalities of trainer recording student training and history reviewing Consequently a trainer can record different exercises for specific users whereas a trainee can perform workouts and review his or her own exercise histories
Date of Award2015 Feb 3
Original languageEnglish
SupervisorWei-Guang Teng (Supervisor)

Cite this