Spatiotemporal gait analysis is well established but remains underused in routine practice. We propose an observer-egocentric approach using smart glasses together with monocular 3D human and camera pose estimation to enable gait analysis during routine clinical observation. To support this, we introduce a benchmark dataset of egocentric video captured by a moving observer, with visual-inertial-based reference spatiotemporal parameters, and use it to evaluate a pipeline for real-world gait analysis. Results show the potential and current limitations of observer-egocentric gait analysis.