The Timed Up and Go (TUG) test is a clinical gold standard for assessing elderly mobility, yet its automated deployment in home-monitoring and resource-limited areas is hindered by high hardware costs and expert calibration requirements. This study introduces a Physics-Data Hybrid framework specifically designed for uncalibrated consumer-grade CMOS cameras, enabling a "plug-and-play" solution for early Parkinson's disease (PD) risk screening. The proposed pipeline integrates learning-based pose perception with a self-evolving physics model to recover absolute metric-scale motion without manual checkerboard calibration. A noise-adaptive fusion strategy is implemented to reconcile 2D pixel dynamics with 3D kinematic consistency, overcoming the inherent scale ambiguity of monocular vision. Crucially, this framework enables the extraction of high-dimensional spatiotemporal parameters-such as stride length coefficient of variation and mean gait velocity-which provide a finer diagnostic resolution for capturing subtle motor fluctuations than conventional timing-only systems. Results from our pilot study with a cohort of 10 subjects demonstrate that these extracted metric features serve as decisive markers for risk staging simulated by dual-task-induced cognitive-motor-interference, achieving 98% screening accuracy and an overall classification accuracy of 87.32%. This framework provides a robust, low-cost tool for ubiquitous telehealth, potentially supporting early PD risk assessment in underserved populations.