Abnormal Walking Gait Analysis Using Silhouette-Masked Flow Histograms

Abnormalities of gait patterns can provide telltale signs of the onset or progression of certain diseases. This paper proposes a simple but effective approach to abnormal gait analysis using computer vision techniques. The proposed method starts with the extraction of human silhouettes from input videos and the computation of frame-to-frame optical flows, then motion metrics based on histogram representations of silhouette-masked flows, and finally gait analysis with eigenspace transformation. Different from current gait classification and recognition studies, the proposed method deals with another interesting problem, namely not only determining different styles of the same walking action but detecting whether or not it is deviated from usual walking pattern, which is expected as a feasible means to deduce physical conditions of people. Experimental results show its promising performance.