The human motion analysis systems have significant importance in today’s scenario across
various industrial domains such as animation, movies, and sports training, as well as in research
areas involving human motor skills, greatly impacting rehabilitation, biomechanics, and diagnostics. However, such systems often come with a high acquisition cost, either due to software
or equipment, as well as logistical issues regarding transportation and implementation outside
laboratory facilities, typically associated with the specificities of the devices used, hindering
access to a larger audience. Bearing this in mind, this work conducts a case study aiming to propose a low-cost alternative using computer vision tools, more precisely, the OpenPose software.
Additionally, digital signal processing techniques are employed to create an analysis system that
achieves satisfactory levels, drastically reducing the number of equipment used and their overall
cost, while maintaining performance comparable to a high-standard system available in the
laboratory dedicated to these studies, the Laboratory of Human Motor Studies (LEMOH), linked
to FFTO-ICS/UFPA. The results of this work have shown promising outcomes in controlled environments (lighting, color contrast, no occurrence of limb occlusion, and structures for equipment
stabilization), achieving correlation coefficients above 85%, as well as a signal-to-noise ratio
(SNR) above 8 dB (decibels) in comparisons made between signals obtained by both systems
concerning the displacement of anatomical points. For comparisons of velocity and acceleration
signals, a correlation coefficient above 80% was obtained, both presenting an SNR value above
4 dB. The estimates of angular joint variation for the tested movement showed average SNR
values above 7 dB.
|
