关键词: 异常行为, 时空马尔科夫随机场, motion 特征, Gibbs能量函数

Abstract: Concerning the sudden-violence abnormal behavior between people, an algorithm which could identify abnormal behavior and other normal behaviors more accurately from images was proposed. This algorithm was based on traditional image segmentation technique, adopted the feature of neighborhood-related-only and the motion feature of adjacent frame in Markov Random Field (MRF), reconstructed the Gibbs energy function. This algorithm considered not only the spatial-information between every pixel and its neighborhood, but also the time-information between successive frames was added in to calculate the whole energy value in the acquired images, which those energy data could be analyzed by using the energy curve. Finally the effectiveness of this approach has been proved by comparing with the traditional optical flow method.

Key words: abnormal behavior, time-spatial MRF, motion feature, Gibbs energy function