Complex action recognition from human motion tracking using wearable sensors
Irvin Hussein López-Nava · LA Referencia (Red Federada de Repositorios Institucionales de Publicaciones Científicas) · 2018
Human motion analysis is defined as any procedure involving any means for obtaining a quantitative or qualitative measure of it. Quantitative analysis involves the measurement of biomechanical variables, such as pressure distribution, temporal and spatial gait parameters, human posture, among others. Calculating biomechanical variables from wearable inertial sensors is possible by using computational techniques for information fusion. From the calculated variables, it is possible to determine the orientation of the limbs, and even recognize certain human actions. This research is about human motion analysis, namely about how human complex actions can be recognised in daily living environments by tracking motion captured by wearable inertial sensors. This is a challenging open problem that involves resources and expertise of various fields, such as digital signal processing or machine learning. The general objective of this research is to propose a method to estimate joint angles of opposite upper and lower limbs using wearable inertial sensors for recognizing complex actions in daily living environments. According to the objectives defined for the present research, the tracking of the opposite upper and lower limbs was carried out using joint angles, and in most of the trials performed according to the proposed first experiment, an error of estimation lower than 10 degrees was achieved. No evidence was found to determine that using estimation of straightforward orientation is better than using estimation of combined orientation to classify the set of complex actions selected in the present research and under the second experimental protocol. Finally, this research has improved the state of the art of the research on human motion analysis based on limited data captured exclusively using wearable sensors, provided that richer information can be extracted from adapted structures such as the kinematic models used in this research.