Running is one of the most popular sports and has tremendous benefits for people's physical and mental health, but unscientific running posture also poses a great risk to people. Monitoring runners' posture, speed, stride frequency and stride length can help reduce their risk of injury during running and also help improve running performance. The monitoring of runners' posture and speed can be achieved by accumulating specific information (e.g., stride frequency and stride length) during running in the runners' daily training program. With this aim, Yuta Suzuki's team at the Center for Urban Health and Sports Research, Osaka City University, Japan, designed a method to estimate foot trajectory and stride length during running using an IMU.

Fig. 1 Installation position and coordinates of IMU

In the past few years, a lot of research work has been conducted in the field of biomechanics using IMUs in gait event monitoring and step length estimation. However, since IMU only measures triaxial linear acceleration, angular velocity and magnetic field strength in its own local coordinate system, it is not possible to estimate foot trajectories and step lengths in the global coordinate system directly from IMU data. And a major problem in calculating trajectories from IMU data is the drift in acceleration and angular velocity measurements, which becomes more and more distorted in the position and orientation assessment as the evaluation time grows. A popular approach to address this drift is to perform a Jetlink integration using the zero-velocity assumption, in which the foot is assumed to have zero velocity at a specific point in the support phase regardless of running speed.

In their study, Yuta Suzuki's team measured the acceleration and angular velocity of the right and left foot using two IMUs mounted on the dorsum of the foot. The foot trajectory and stride length were estimated with the assumption of zero velocity for more IMU data, and the rotation of the IMU was estimated to calculate the displacement in the medial and lateral directions and vertical directions in the middle of the phase of two consecutive gait supports. Specifically, the authors propose two algorithms, Spatial Error Correcting (SEC) and Linear Dedrifting (LD), to perform data calibration and then compare the calculated footprint and stride length to the motion capture data.

Figure 2 Gait diagram

In this study, 79 runners performed running trials at 5 different speeds, and a total of 389 trials and 1414 strides (699 strides for the left foot and 715 strides for the right foot) were obtained by IMU and VICON, with a mean running speed of 3.42 ± 0.72 m/s and a range of variation from 1.86 to 5.89 m/s. As shown in Figure 3, the LD and SEC methods were used to obtain the right foot medial (X), anterior-posterior (Y) and vertical trajectories (Z) and compared with the reference values. As shown, the correlation coefficients of Y-coordinate are very high using both LD and SEC methods, and the correlation coefficient of Z-coordinate for SEC method is larger than that of LD method. In addition, the root-mean-square error of the estimated foot trajectory of the SEC method is smaller than that of the LD method compared with the reference value.

Figure 3 Schematic diagram of the foot trajectory obtained by LD and SEC

In summary, the algorithm proposed by the team may be able to provide more accurate estimates of gait parameters for runners.