*5.2. Accuracy of the Kinematic Parameter Evaluation*

The measurement accuracy of Microsoft Kinect v2 in clinical estimation of motor functions [27,28,30] and body CoM [18,19,25,27,29] has been previously assessed: this was confirmed also by our experiment. The comparison of the parameter measurement respect to the gold reference system cannot be performed directly because the Kinect skeleton model and the optoelectronic marker set have different body landmark positions. Furthermore, for every parameter, we want to estimate an average accuracy based on all the trials per task acquired.

Then, for every essential parameter *i*, characterizing one or more tasks, the parameter samples of each associated task trials were joined together into a single parameter sample sequence (PSSi). The PSSi sequences measured by the two systems were then compared by evaluating the Pearson correlation coefficient *ri*. In Table 4, the *ri* coefficients indicate a significant correlation that ranges from good to strong for all the examined parameters. In Figures 6 and 7 are shown two examples of the ANGKNEE and the ANGTRUNK variations for a LA and an AC task trial. In Figure 6, it is interesting to point out that the last movement is characterized by a significant reduction in amplitude and duration. In Figure 7, the presence of a secondary peak indicates that the PD subject had an instability event at the corresponding time. These anomalies are hardly identified by neurologists.


**Table 4.** Mean and standard deviation of the Pearson's correlation coefficients for essential parameters estimated by the two systems.

**Figure 6.** Example of the ANGKNEE variations during the LA task performance of a PD subject: the last movement at 8.9 s is characterized by significant reduction in both amplitude and duration.

**Figure 7.** Example of the ANGTRUNK variations during the AC task performance: the secondary peak at 8.5 s indicates the presence of an instability event in the final standing stance.
