A Quantitative Model To Evaluate Wrist-Rotation In Golf (P5)
Quantitative Analysis for Wrist Rotation
In this section, we investigate quality measure with respect to the wrist rotation. We ﬁnd a linear projection of feature space that monotonically changes while the angle of rotation varies. This projection is obtained by LDA and is fed to a regression model to quantify the degree of improperness of movement. The system requires a dataset with trials obtained from several variations of the wrist rotation. Swings performed for each variation are required to account for consistent physical movements. That is, within each group gi, the trials should have the same degree of wrist rotation while different groups exclusively differ with respect to the rotation. This would imply the need for a highly controlled experimental environment. Therefore, we use a home swing trainer which helps to maintain an in-plane swing. The device has a rigid rod with one end mounted on the wall and the other end on which the golf club slides as shown in Fig. 7. The rod and the club are connected at one end as illustrated in Fig. 7.a where the club can rotate along its axis. We modiﬁed the club to ensure that it was ﬁxed about its axis during every trial. This makes our experiments even more controlled because the amount of wrist rotation remains consistent provided that the subject maintains a steady grip through out each trial. The club is marked at both head and grip sides to depict the angle of rotation. Hence, by placing sensor nodes at locations associated with angles, a consistent angular wrist rotation is maintained.
An extremely important requirement in building our model is to restrict each trial to a speciﬁc type of bad swing. The experimental setup for the wrist rotation, as shown in Fig. 7.b, keeps all segments of a swing within the same plane resulting in maintaining an inplane swing. This ensures that different trials can differ only in the amount of the wrist rotation. The resulting model then will be able to quantify incorrect swings in terms of angular rotation. In reality, however, several types of mistakes can be made independently by the golfer, each of which must be quantiﬁed using individual models. Integrating evaluation of mistakes other than wrist rotation into our existing model is a problem that we will investigate in the future.
To build the quantitative model for wrist rotation, N number of observations associated with k different variations of wrist rotation are required. The data collected for k angles form a dataset of groups g1,g2,…,gk. The LDA projection can be derived as described previously. Finally, the parameters of a linear regression are calculated based on N values of the discriminant functions introduced by LDA.