Human behavior detection method with direction change invariant features

In this paper, we propose a new feature homogenization method using cubic higher-order local auto-correlation (CHLAC) to detect changes in human behavior. Conventional human behavior detection using CHLAC exhibits a high level of performance, but has difficulty in distinguishing between abnormal and normal movement. We propose a method with improved handling and statistical processing of mask patterns to suppress the change in the amount of features according to the direction of movement of the person. This provides a robust method of detecting changes in direction. A computer simulation using the proposed method demonstrates a superior performance composed to a conventional method in the recognition of abnormal human behavior.