A New View-Calibrated Approach for Abnormal Gait Detection

Gait, or the style of walking, has been recently a popular topic in vision-based analysis. Most vision-based works about gait are devoted to the application of human recognition, but abnormal walking styles are rarely discussed. Accordingly, a vision-based method is proposed to analyze abnormal types of walking. In the proposed method, a background subtraction algorithm is applied to segment out the silhouette of the walker at each frame in a sequence. For each frame, we define a feature based on the length between two legs and the height of the individual, called aspect ratio (AR). By observing this feature value across time (or frame), a periodic wave is obtained. With this analysis, a few abnormal types of walking can be distinguished. Since an oblique camera view angle causes a distortion of the AR wave, a rectification mechanism is proposed based on a camera pinhole model to reduce the view angle effect. The experimental results show that the proposed rectification method identified abnormal walking patterns reliably irrespective of the direction in which the individual walks relative to the camera plane.