Parkinson’s disease is a progressive neurodegenerative disorder characterized by motor impairments, among which bradykinesia represents one of the most prominent and clinically relevant symptoms. The assessment of bradykinesia is typically performed through clinical observation and rating scales, which can be subjective and provide only intermittent measurements. In this context, automated video-based analysis offers a promising solution for objective, continuous, and non-invasive monitoring of disease severity. The objective of this thesis is the development of a lightweight deep learning algorithm for real-time estimation of bradykinesia severity from facial video data, with a specific focus on deployment in vehicle-related applications. The proposed approach aims to balance high predictive performance with low computational complexity, enabling real-time processing on low-power devices.