With the global population aging, promoting healthy aging has become a critical societal objective. Movement disorders, which include age-related motor decline and neurodegenerative diseases such as Parkinson's disease, significantly impair quality of life and impose substantial healthcare burdens. Early detection and intervention are crucial, yet current assessment methods primarily rely on subjective questionnaires and physical examinations, which are inefficient, resource-intensive, and lack standardization. To address these challenges, this study proposes a novel movement disorder assessment algorithm that leverages object detection, pose estimation, and action recognition techniques. By exploiting the differences in gait-related stability, coordination, and muscle activity between individuals with movement disorders and healthy individuals, the proposed algorithm employs a two-stage approach: (1) a keypoint extraction algorithm composed of the object detection algorithm and the pose estimation algorithm and (2) an improved action recognition algorithm based on the spatial-temporal graph convolutional network (ST-GCN), which incorporates a data-dependent adjacency matrix, multi-scale temporal window transformation, multimodal aggregation, and contrastive learning for precise classification. Experimental results show a 10.24% accuracy improvement over ST-GCN, achieving an accuracy of 82.03%. This method offers a more efficient, convenient, and scalable alternative to traditional approaches, providing a valuable foundation for intelligent elderly care and future research in movement disorder diagnostics.