Single Camera Based Gait Analysis Method with Scaled and Image Coordinate Key-points

Clinical experiments and scientific evidence show the relations between gait and cognition. Neurological diseases (like stroke and cerebral palsy) distract patients' attention and restrict their executive functions. Furthermore, these diseases impair patients' movement and limit their social participation. Quantitative motion assessment is significant to medical decision-making for neurological disorders but is possible only with expensive motion capture systems and professional medical workers. Here, an end-to-end video-based quantitative gait analysis algorithm is proposed. Unlike previous works, which only focus on image coordinate key-points, our algorithm also utilizes scaled key-points that can depict the size of patients. These two groups of features are sent to other neural networks to extract temporal and spatial characteristics, and then the deep features are fused by the attention mechanism. Finally, a statistic model is used to process the proposed outputs to get a final result. Experiments on the public patients' dataset show the effectiveness and accuracy of our model. Furthermore, the ablation study proves that every step of the proposed method contributes to the whole system's performance.