Abstract
Basketball activity classification can help document players’ statistics,
allow coaches, trainers and the medical team to quantitatively supervise
players’ physical exertion and optimize training strategy, and further
help prevent potential injuries. Traditionally, sports activity classification
was done by manual notational, or through multi-camera systems or motion sensing
technology. These methods were often erroneous and limited by space. This study
presents a basketball activity classification model based on Dynamic Time
Warping (DTW) and body kinematic measures. Twenty participants, including 10
experienced players and 10 novice players, were involved in an experimental
study. The experienced and novice players differed in their years of playing
basketball. Four basketball movements, including shooting, passing, dribbling,
and lay-up were classified by kinematic measures. The results indicate that the
proposed model can successfully classify different basketball movements with
high accuracy and efficiency. Specifically, with the resultant acceleration of
the hand, this model can achieve classification precision, recall, and
specificity up to 0.984, 0.983 and 0.994, respectively. Findings from this study
supported the feasibility of using DTW in real-time sports activity
classification and provided insights into the optimal sensor placement for
basketball activity classification applications.
Key words
sports activity classification - dynamic time warping - upper body kinematics - basketball
