Gait measurement using Radial Basis Function Networks and comparison of its performance with analytical methods

In this study, Radial Basis Function Networks (RBFN), a neural networks structure, is investigated to employ in sensor based motion measurement system of an electronic above knee prosthesis that ensures above knee amputees walk at varying speed. Having gait measurements of a custom designed image based measurement system as a reference, motion measurement performance of RBFN is compared with the performance of analytical methods on basis of knee angle measurements. Accuracy of knee angle estimations of RBFN is observed to be better than the accuracy of analytical methods using only accelerometers, other one using combination of accelerometer and gyroscope and another one called virtual sensor method. One advantage over analytical methods that RBFN offers is the possibility of decreasing number of sensors required to estimate knee angle. Even in case of single accelerometer attached on shank where lowest performance in knee angle estimation is observed, accuracy of RBFN is higher than the one analytical methods provide. From results of experiments, it is found that gyroscopes affects performance more than accelerometers do in knee angles estimation of RBFN.