Fig. 2

Speed estimation from accelerometer, gyroscope, and behavioural annotation data. We developed an algorithm to estimate locomotion speed from IMU data and behavioural annotation. We used the behavioural annotation in our power calibration dataset to select locomotion bouts preceded and followed by static behaviour (resting or vigilance). We refined the start and end of these locomotion bouts by using acceleration and angular velocity thresholds to ensure that starting and ending velocity and acceleration were zero. We removed gravitational acceleration by fusing accelerometer and gyroscope data in an internal error-state Kalman filter, and rotated gravity-compensated acceleration from the sensor frame to the global coordinate frame using quaternions obtained by applying the function imufilter in MATLAB. We integrated acceleration in the global coordinate frame to obtain velocity and removed integration drift through linear compensation using the information that initial and final velocity are zero. Finally, we computed the vectorial norm of the velocity components in the horizontal plane to obtain the instantaneous locomotion speed signal. Shown here is a bout of running preceded and followed by vigilance