Deep-learning-based approaches have begun with 2D pose estimation, which automatically estimates human joint centers from 2D RGB images, outputting the 2D coordinates in the images ( Toshev and Szegedy, 2014 Wei et al., 2016 Papandreou et al., 2018). When compared to the approach using RGB-Depth cameras such as Kinect ( Clark et al., 2012 Pfister et al., 2014 Schmitz et al., 2014 Gao et al., 2015), a deep-learning-based approach has less constraints on the distance between the camera and the target to be measured as well as the sampling rate of video recording. Most of these algorithms train the neural network using manually labeled image data and then estimate the human posture, such as joint centers and skeletons, when the user inputs the images or videos to the trained network. Recently, automatic human pose estimation using deep learning techniques have attracted attention amongst computer vision researchers. Therefore, it is desirable to many biomechanics researchers to develop a markerless motion capture that is easy to use for measurement. Markerless measurements without such environmental constraints can facilitate new understanding about human movements ( Mündermann et al., 2006) however, complex information processing technology is required to make an algorithm that recognizes human poses or skeletons from images. For example, measurements cannot be performed in environments wherein wearing markers during the activity is difficult (such as sporting games). However, traditional marker-based approaches have significant environmental constraints. Motion capture systems have been used extensively as a fundamental technology within biomechanics research. In conclusion, this study demonstrates that, if an algorithm that corrects all apparently wrong tracking can be incorporated into the system, OpenPose-based markerless motion capture can be used for human movement science with an accuracy of 30 mm or less. The primary reason for mean absolute errors exceeding 40 mm was that OpenPose failed to track the participant's pose in 2D images owing to failures, such as recognition of an object as a human body segment or replacing one segment with another depending on the image of each frame. Quantitatively, of all the mean absolute errors calculated, approximately 47% were 40 mm. The results demonstrated that, qualitatively, 3D pose estimation using markerless motion capture could correctly reproduce the movements of participants. The differences in corresponding joint positions, estimated from the two different methods throughout the analysis, were presented as a mean absolute error (MAE). Participants performed three motor tasks (walking, countermovement jumping, and ball throwing), and these movements measured using both marker-based optical motion capture and OpenPose-based markerless motion capture. This study aims to develop a 3D markerless motion capture technique, using OpenPose with multiple synchronized video cameras, and examine its accuracy in comparison with optical marker-based motion capture. There is a need within human movement sciences for a markerless motion capture system, which is easy to use and sufficiently accurate to evaluate motor performance.
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |