Integrated 2D video-based analysis using DeepLabCut pose estimation
Readily available pose estimation algorithms often require extensive calibration methods for 3D pose estimation whereas 2D pose estima- tion does not always capture all motion. At present, 3D pose estima- tion approaches often require non-trivial deep learning methods and there is a lack of su cient 3D datasets. In this study, the feasibility of integrating two di erent camera views of human locomotion using DeepLabCut was investigated. A signi cant perspective di erence in most x-components was observed, while all y-components did not have a signi cant perspective di erence. This suggests that integrating 2D pose estimation is not feasible using this approach.
Faculteit der Sociale Wetenschappen