EuclideanPose: Novel Boolean Filtering Approach for Accurate Multiview 3D Human Pose Estimation Based on Euclidean Distance

Motion capture technology has advanced significantly, capturing real-time motion imagery in 2D and 3D. While marker-based systems have achieved sub-millimeter accuracy, their setup complexity and interference with natural movement drive the search for alternative methods. Wearable devices like IMUs overcome some limitations but still disrupt natural movement and suffer from drift. Marker-less motion capture., particularly 3D human pose estimation using multi-view 2D estimations, presents a promising alternative. However, methods like Weighted Direct Linear Transform (WDLT) depend heavily on visibility weights from 2D pose estimation, which can be unreliable. This study introduces the EuclideanPose method, a novel approach using boolean filtering based on Euclidean distance to eliminate less accurate 2D estimations, enhancing 3D reconstruction accuracy. Experimental results demonstrate that EuclideanPose achieves an average error reduction in hip and knee angles in gait cycle to 1.68° and 2.32° respectively, compared to WDLT's 2.25° and 3.31°.