ESTIMATION OF ENERGY EXPENDITURE DURING WALKING USING A KINEMATIC SENSOR

ESTIMATION OF ENERGY EXPENDITURE DURING WALKING USING A KINEMATIC SENSOR O. Genton1, B. Najafi1, E. Tam2, C. Moya2, G. Ferretti2, K. Aminian1 1 Swiss Federal Institute of Technology EPFL- LMAM, Station 11, CH-1015 Lausanne 2 Neurosciences Fondamentales, Centre Médical Universitaire, CH-1211 Genève 4 *** INTRODUCTION Energy expenditure during walking depends on subject’s weight, walking speed and walking path incline. In this study, we propose a new method to measure the walking speed and the incline using a kinematic sensor. The estimated values for speed and incline were then used to predict the energy expenditure. METHODS Fifteen healthy subjects (6 women - 9 men, 27±6 years, 175.5±10.5 cm, 73.5±16.5 kg (m)) participated in this study. Each subject were asked to perform 14 walking tests on a treadmill (incline : -15%, -10%, 0%, 5%, 10% and 15%, speed : 2.5 to 5.0 Km/h) while carrying a sensor module consisting of three accelerometers and a gyroscope fixed on the heel (Physilog®) and a reference device to measure energy expenditure (SensorMedics VMAX29®). The period of foot-flat was first detected according to the foot’s kinematics. The incline (s) and walking speed (v) were then estimated by integration of the foot acceleration (vertical and horizontal) between each foot-flat cycle. A model based on a multiple non-linear regression was developed to estimate energy expenditure. This model was elaborated based on the estimated data as well as reference values from the first half of the test persons and then was validated on the second half. RESULTS The inertial sensor enables recognizing the incline with a mean error of 0.05 ± 0.94% (maximum 0.6 ± 2.1%) and the speed with an average error of 0.004 ± 0.2 km/h (maximum 0.22 ± 0.17 km/h). Among 14 different criteria used for the statistical model, the most important predicting factors for estimating the energy expenditure were s, v, v 2 , m ⋅ v 2 . Using this statistical model, a relatively high correlation was observed between the estimated and the reference values (r>0.90). The mean error of the predicted energy expenditure for the sub-group used for the validation was 4.0 ± 11.1% over all activities on the positive inclines. DISCUSSION The developed methods in this pilot study enable estimating the speed and the incline for each walking cycle using an ambulatory device. We showed also that a resolution of a few cycles could be sufficient to confirm the incline as well as the speed and to predict the energy expenditure for the positive inclines. However, for the negative inclines, a different model should be developed. CONCLUSION Although the number of subjects was limited, this pilot study reveals a promising method to estimate energy expenditure when walking using a simple ambulatory device. This device can be worn during a long period in free condition without hindering the usual activities of the subject.