Parkinson disease, a neurodegenerative disorder that deteriorates over time, slowly impacts 1% of the elderly population worldwide. The severity assessment is crucial to plan medication and treatment plans effectively. Current procedures for assessing severity include manual visual assessment of motor function using tests like finger tapping. This assessment and determination of the severity based on clinical rating scale may vary between clinicians based on their expertise. This paper proposes a computer vision-based automated severity prediction from finger tapping (FT) test. The video recording containing the FT test on both hands is analyzed using pose estimation algorithms. Movement features associated with bradykinesia were computed from video recordings, and machine learning classification algorithms were implemented to characterize disease severity. The results of proposed approach showed higher accuracy of over 90% in differentiating levels of severity using decision tree classification algorithm.