Inertial Sensor Error Reduction through Calibration and Sensor Fusion

Autor(es)

Lambrecht, Stefan ^{[1, 2, 3]} ; Nogueira, Samuel L. ^[4] ; Bortole, Magdo ^[1] ; Siqueira, Adriano A. G. ^[5] ; Terra, Marco H. ^[6] ; Rocon, Eduardo ^[7] ; Pons, Jose L. ^[1]

Afiliação do(s) autor(es)

^[1] CSIC, Neural Rehabil Grp, Av Dr Arce 37, Madrid 28002 - Spain

^[2] Katholieke Univ Leuven, Dept Mech Engn, Div PMA, Celestijnenlaan 300B, B-3001 Heverlee - Belgium

^[3] Katholieke Univ Leuven, Dept Biomed Kinesiol, Tervuursevest 101, B-3001 Heverlee - Belgium

^[4] Univ Fed Sao Carlos, Dept Elect Engn, BR-13565905 Sao Paulo - Brazil

^[5] Univ Sao Paulo, Ctr Robot Sao Carlos, Dept Mech Engn, BR-13565905 Sao Paulo - Brazil

^[6] Univ Sao Paulo, Dept Elect Engn, Ctr Robot Sao Carlos, BR-13565905 Sao Paulo - Brazil

^[7] CSIC, Grp Neural & Cognit Engn, Ctra Campo Real Km 0-200, Arganda Del Rey 28500 - Spain

• Tipo de documento: Artigo Científico
• Fonte: SENSORS; v. 16, n. 2 FEB 2016.
• Citações Web of Science: 6

Resumo

This paper presents the comparison between cooperative and local Kalman Filters (KF) for estimating the absolute segment angle, under two calibration conditions. A simplified calibration, that can be replicated in most laboratories; and a complex calibration, similar to that applied by commercial vendors. The cooperative filters use information from either all inertial sensors attached to the body, Matricial KF; or use information from the inertial sensors and the potentiometers of an exoskeleton, Markovian KF. A one minute walking trial of a subject walking with a 6-DoF exoskeleton was used to assess the absolute segment angle of the trunk, thigh, shank, and foot. The results indicate that regardless of the segment and filter applied, the more complex calibration always results in a significantly better performance compared to the simplified calibration. The interaction between filter and calibration suggests that when the quality of the calibration is unknown the Markovian KF is recommended. Applying the complex calibration, the Matricial and Markovian KF perform similarly, with average RMSE below 1.22 degrees. Cooperative KFs perform better or at least equally good as Local KF, we therefore recommend to use cooperative KFs instead of local KFs for control or analysis of walking. (AU)

• Processo FAPESP: 12/05552-9 - Sistema Robusto de Estimativa de Posicionamentos Angulares Absolutos e Forças de Interação para Exoesqueletos de Membros Inferiores
• Beneficiário: Samuel Lourenço Nogueira
• Modalidade de apoio: Bolsas no Brasil - Doutorado