Dominant pattern extraction from 3-D kinematic data

• Published in: IEEE transactions on biomedical engineering Vol. 46; no. 1; pp. 100 - 106
• Main Authors: Stokes, V.P., Lanshammar, H., Thorstensson, A.
• Format: Journal Article
• Language: English
• Published: New York, NY IEEE 01-01-1999; Institute of Electrical and Electronics Engineers
• Subjects: Algorithms; Bioinformatics; Biological and medical sciences; Biomechanical Phenomena; Biomechanics; Data mining; Exercise equipment; Feature extraction; Gait - physiology; Harmonic analysis; Humans; Investigative techniques, diagnostic techniques (general aspects); Kinematics; Locomotion - physiology; Mathematical transformations; Medical sciences; Medicin och hälsovetenskap; Motor drives; Neuroscience; Osteoarticular system. Muscles; Pathology. Cytology. Biochemistry. Spectrometry. Miscellaneous investigative techniques; Pattern analysis; Signal Processing, Computer-Assisted; Signal to noise ratio; Biomechanics; Locomotion; Data analysis; Stride; Gait; Kinematics; Three dimensional system; Information extraction; Repetition; Cyclic system; SVD; SINGULAR-VALUE DECOMPOSITION; stride-cycle; HUMAN LOCOMOTION; locomotion; SIGNALS; kinematics; trajectory; PREDICTION; three-dimensional (3-D); gait; HUMAN GAIT; PELVIS; THORAX; human; treadmill
• ISSN: 0018-9294; 1558-2531; 1558-2531
• DOI: 10.1109/10.736764

A new method for the extraction of a repeating pattern in cyclic biomechanical data is proposed-singular value decomposition pattern analysis (SVDPA). This method is based on the recent work of Kanjilal and Palit (1994, 1995) and can be applied to both contiguous and repeated trials without being constrained to be strictly periodic. SVDPA is a data-driven approach that does not use a preselected set of basis functions; but instead utilizes a data matrix with a special structure to identify repeating patterns. Several important features of SVDPA are described including its close relationship to the Kahunen-Loeve transform. The dominant pattern is defined as the average energy component (AEC). The AEC is obtained from the SVD of the data matrix and is equivalent to the optimal [maximal signal-to-noise ratio (SNR)] ensemble average pattern. The degree of periodicity and SNR for the AEC are defined explicitly from the singular values of the data matrix. The authors illustrate the usefulness of SVDPA for dominant pattern extraction by applying it to the quasiperiodic three-dimensional trajectory of a marker attached to the trunk during treadmill locomotion. The AEC obtained for the normalized trajectory and error estimates at each point suggests that SVDPA could be a useful tool for the extraction of the fine details from cyclic biomechanical data.