Validation of running speed across different speeds and slopes using Kinetyx sensory insoles

Authors: Eric C. Honert¹, Sandro R. Nigg¹, Benno M. Nigg¹, Sam Blades²

1 University of Calgary
2 Kinetyx Sciences Inc.

Highlights 

- The novel Kinetyx sensory insole algorithm estimates running speed better than an IMU-only based algorithm. 

- First study to evaluate running speed algorithms across different running slopes. 

Background 

Running is a popular sport and means of exercise. Many runners utilize technology such as sports or smart watches in order to gain a better understanding of their performance and to monitor their training. Devices that rely on global positioning systems (GPS) have the ability to track runners’ position and speed in real time; however, such devices have several deficits. Sports watches typically record speed once per second [1], [2], which is longer than a running stride, and GPS tracking accuracy can vary based location and weather (e.g. forested and cloudy, [3]). Alternatively, sensors such as inertial measurement units (IMUs) can provide running specific metrics such as running speed [4] on a stride-by-stride basis regardless of the location or weather. A further benefit to IMU sensors is they consume less power, meaning they can be used for a longer duration. IMU running speed algorithms fuse data from accelerometers and gyroscopes within the IMU in order to estimate running speed. Yet, such speed estimates have only been validated on level ground [4]–[8], despite the variety of terrain that runners encounter during training and competition [9]. Further, although IMU-based running speed measures are more versatile than GPS-based measures, they can still have speed estimation errors between 3.5-6.9% [4]. Improvements in these running speed algorithms could be garnered by fusing information from plantar pressure data. Thus, the purpose of this white paper was two-fold. The first objective was to evaluate running speed computed from novel algorithm that fuses data from IMU and plantar pressure sensors that are within the Kinetyx sensory insole. This new algorithm was then compared to an algorithm that only uses IMU data to compute running speed. The second objective was to evaluate and compare these two algorithms across different running slopes. 

Methods 

Eighteen recreationally active subjects provided written informed consent and participated in this study. The protocol was approved by the University of Calgary’s Conjoint Health Research Ethics Board (REB20- 1734). Subjects ran on an instrumented treadmill (Bertec, Columbus, OH, USA) on level ground followed

by uphill and downhill on a 10.5% grade. Subjects ran on level ground at 9.4, 10.8, 12.2 and 13.7 km/hr, uphill at 9.4, 10.1 and 10.8 km/hr, and downhill at 9.4, 10.1 and 10.8, and 12.2 km/hr for one minute each. Subjects were not requested to run at faster speeds during non-level ground trials to avoid the potential of injury. Data were collected from the Kinetyx sensory insole, which includes IMU and plantar pressure sensors. 

Running speed was then computed using two different algorithms. First, a novel running speed algorithm was developed using all sensors on Kinetyx sensory insole based on previously published, IMU-based, running speed algorithms [4], [10]. Hereafter, this algorithm will be referred to as the Kinetyx sensory insole algorithm. Next, an IMU-only running speed algorithm was created based on previous literature [4], [10]. This algorithm was created in order to draw comparisons between the IMU-only approach and Kinetyx sensory insole algorithm. The mean absolute percent errors for each running speed and slope were then computed to evaluate both running speeds algorithms against the treadmill belt speed. The running speed percent errors for both algorithms were then statistically compared with paired t-tests. 

Results

The novel Kinetyx sensory insole algorithm estimated running speed (Fig. 1) across many different speeds and slopes for over 21,000 strides. This algorithm produced an average error below 4.3% (Fig. 2). For all speeds and slopes, the Kinetyx sensory insole reduced the average percent error as compared to the IMU-only algorithm by up to 29% (p < 0.001, Fig. 2). For individual subjects, this error was reduced up to 54%. 

Summary

Regardless of the speed or slope, the novel Kinetyx sensory insole algorithm had a lower running speed error than the IMU-only algorithm. Additionally, this algorithm had an average error of 3.0% on level ground, which was better than IMU-only based algorithms (3.5 - 6.9%) [4]. This study also represents one of the first validations of running speed algorithms across different speeds and slopes, as the majority of previously published IMU running speed algorithms have focused on level ground running [4]–[8]. In total, the implementation of the Kinetyx sensory insole algorithm would allow runners to better track their workouts and overall performance. Accurate running speed estimates can also be used to compute additional biomechanical metrics such as ground reaction forces [11] and running power [12].

References

[1] S. Bichler, G. Ogris, V. Kremser, F. Schwab, S. Knott, and A. Baca, “Towards high-precision IMU/ GPS-based stride-parameter determination in an outdoor runners’ scenario,” Procedia Eng., vol. 34, pp. 592–597, Jan. 2012, doi: 10.1016/j.proeng.2012.04.101. 

[2] C. J. de Ruiter, B. van Oeveren, A. Francke, P. Zijlstra, and J. H. van Dieen, “Running Speed Can Be Predicted from Foot Contact Time during Outdoor over Ground Running,” PLOS ONE, vol. 11, no. 9, p. e0163023, Sep. 2016, doi: 10.1371/journal.pone.0163023. 

[3] R. Gilgen-Ammann, T. Schweizer, and T. Wyss, “Accuracy of Distance Recordings in Eight Positioning-Enabled Sport Watches: Instrument Validation Study,” JMIR MHealth UHealth, vol. 8, no. 6, p. e17118, Jun. 2020, doi: 10.2196/17118. 

[4] M. Zrenner, S. Gradl, U. Jensen, M. Ullrich, and B. M. Eskofier, “Comparison of Different Algorithms for Calculating Velocity and Stride Length in Running Using Inertial Measurement Units,” Sensors, vol. 18, no. 12, Nov. 2018, doi: 10.3390/s18124194. 

[5] G. P. Bailey and R. Harle, “Assessment of Foot Kinematics During Steady State Running Using a Foot-mounted IMU,” Procedia Eng., vol. 72, pp. 32–37, 2014, doi: 10.1016/j.proeng.2014.06.009. 

[6] D.-K. Chew, K. J.-H. Ngoh, D. Gouwanda, and A. A. Gopalai, “Estimating running spatial and temporal parameters using an inertial sensor,” Sports Eng., vol. 21, no. 2, pp. 115–122, Jun. 2018, doi: 10.1007/ s12283-017-0255-9. 

[7] M. Falbriard, A. Soltani, and K. Aminian, “Running Speed Estimation Using Shoe-Worn Inertial Sensors: Direct Integration, Linear, and Personalized Model,” Front. Sports Act. Living, vol. 3, p. 29, 2021, doi: 10.3389/fspor.2021.585809. 

[8] M. V. Potter, L. V. Ojeda, N. C. Perkins, and S. M. Cain, “Effect of IMU Design on IMU-Derived Stride Metrics for Running,” Sensors, vol. 19, no. 11, Art. no. 11, Jan. 2019, doi: 10.3390/s19112601. 

[9] A. D. Townshend, C. J. Worringham, and I. B. Stewart, “Spontaneous Pacing during Overground Hill Running,” Med. Sci. Sports Exerc., vol. 42, no. 1, pp. 160–169, Jan. 2010, doi: 10.1249/ MSS.0b013e3181af21e2. 

[10] A. M. Sabatini, C. Martelloni, S. Scapellato, and F. Cavallo, “Assessment of walking features from foot inertial sensing,” IEEE Trans. Biomed. Eng., vol. 52, no. 3, pp. 486–494, Mar. 2005, doi: 10.1109/ TBME.2004.840727.

[11] R. S. Alcantara, W. B. Edwards, G. Y. Millet, and A. M. Grabowski, “Predicting continuous ground reaction forces from accelerometers during uphill and downhill running: A recurrent neural network solution,” Jul. 2021. doi: 10.1101/2021.03.17.435901. 

[12] V. Cerezuela-Espejo, A. Hernández-Belmonte, J. Courel-Ibáñez, E. Conesa-Ros, A. Martínez-Cava, and J. G. Pallarés, “Running power meters and theoretical models based on laws of physics: Effects of environments and running conditions,” Physiol. Behav., vol. 223, p. 112972, Sep. 2020, doi: 10.1016/j.physbeh.2020.112972.