Leg Joints Angle Estimation During Walking Using the Motion of the Posterior Superior Illiac or Greater Trochanter Points

Estimation of the trajectories of the leg’s joints is of importance in gait studies, as well as in the design of motion planners and high-level controllers for exoskeletons, orthotics, prosthetics, and humanoid robots. Human locomotion is a harmonic phenomenon which benefits from collaboration between different parts of the leg. This collaboration, together with taking into account the natural hierarchy in the human body structure, necessitates paying attention to the fact that the motions of the legs’ lower limbs are influenced by the motions of the upper ones. Having this point and its potential consequences in mind, this study aims to create a relationship between the legs’ joints, and the motion of the posterior superior illiac (PSI) or great trochanter (GTR) points, separately. From anatomical point of view, both of the points are above the ankle, knee, and hip joints. To continuously map the inputs to the outputs, without requiring switching rules, speed estimation, gait percent identification or look-up tables, a nonlinear auto-regressive modeling with wavelets and neural network is used. The proposed approach is investigated for forty-two subjects at different walking speeds. The method is tested for six case studies, in which their root mean square (RMS) errors, mean absolute errors (MAEs) and correlation coefficients
$\rho _{cc}$
are compared. The results show that using GTR point leads to higher estimation accuracy. For instance, in one of the testing case studies,
$\rho _{cc}$
were 0.97, 0.95, 0.91 using GTR point, in comparison to 0.95, 0.93, 0.87 using PSI point, for the hip, knee, and ankle joints, respectively. A similar trend was also observed for root mean squared errors (RMSE) and mean absolute errors (MAEs). In addition, it is found that highest performance occurs in hip angles estimation, and least performance is seen for the ankle joint. Furthermore, the impact of using both velocity and acceleration inputs on the estimation accuracy is also investigated. The results show that using velocity or acceleration of the GTR or PSI inputs leads to relatively similar results. Nonetheless, the results related to the GTR point are in general better. The impacts of using both velocity and acceleration inputs as well as different estimator functions (such as sigmoid function) are also investigated and discussed.