*4.1. Comparison to Existing Literature*

Different papers already evaluated the stride length or the velocity of single strides. Three of these papers are listed in Table 5. These three publications used similar approaches to ours: Bailey et al. [12] used a trajectory approach using a linear dedrifting technique; Gradl et al. [6] used the described acceleration approach; and Hannink et al. [16] a DCNN approach.


**Table 5.** Results of other publications related to stride length and velocity calculation.

With respect to the standard deviation, the results of the trajectory implementation of Bailey et al. [12] are better than our results (Table 4). They also evaluated 1800 strides; however, these strides only originated from running velocities ranging from 2.3–3.4 m/s. We also evaluated our results for this velocity range and obtained an error of 0.004 ± 0.107 m/s. We observe that our standard deviation is still higher than the standard deviation reported from Bailey et al. One reason for that might be the higher number of different runners with different running styles who participated in our study. Furthermore, their study was conducted on a treadmill. On a treadmill, the variability of different strides at a given velocity is lower and does not reproduce overground running kinematics [39].

The errors reported by Gradl et al. [6] were obtained on a smaller database than the one presented in this paper. Thus, our worse results are due to the higher variability in our dataset, which the second degree polynomial could not appropriately approximate.

The results of Hannink et al. [16] were evaluated for gait in geriatric patients. Hence, there is a general difference in the stride patterns, causing differences in the results. Further differences between the setup of our network architecture and study population are listed and discussed in the following section.
