A comparison of footfall detection algorithms from walker mounted sensors data

Footfall detection is an essential element in the quantification of human walking patterns. It is often used for gait analysis and for the assessment of gait disorders. Most current methods are either subjective and rely on human observation, or are performed using complex and expensive equipment, and both solution types are primarily used in laboratory settings. We address the aforementioned systems limitations using low cost sensors mounted on an instrumented walker. The sensors namely distance encoders, force sensors and an accelerometer, acquire kinematic signals. We propose a new footfall extraction algorithm based on the accelerometer z-axis signal and compare it to a previously proposed algorithm which was based on force sensors. Both algorithms make use of Empirical Mode Decomposition (EMD) in order to decompose, filter and reconstruct the respective kinematic signals. Subsequently, threshold based peak detection is applied to estimate potential footfalls. The algorithms results are validated using video of subjects carrying out walking tests. The best detection accuracy for both algorithms was achieved when reconstructing the decomposed signal from the 3rd Intrinsic Mode Function level of the EMD signal. The algorithm using the accelerometer signal demonstrated greater detection accuracy of 86%, whereas the force sensor algorithm yielded an accuracy of 69%. The results imply that combination of the simple low cost accelerometer mounted on a walker and the new footfall detection algorithm, may provide a useful and affordable method of gait analysis.