基于整体能量模型的异常行为检测

为了对少数人群间的突发异常行为进行检测,提出一种基于整体能量模型表示的方法来较准确地辨识正常行为中的异常行为。该整体能量特征算法不仅包括了行为发生双方的个体能量特征,还包括行为发生双方的交互能量特征,这两个因素构成异常行为的整体能量特征。而能量特征的基本参数则是在光流场特征的基础上增加了运动角度变化等特征来深入反映行为的激烈性。最后在不同行为类型视频上测试该算法,并与传统光流能量算法进行测试对比,实验结果展示了整体能量模型相比于传统光流能量算法的高效性以及准确性,能够达到预期目的。

Abnormal behavior detection based on globe energy model

Concerning the sudden-violence abnormal behavior between people, this paper proposed an algorithm which couldidentify the abnormal behavior from other normal behavior more accurately. This globe energy not only included the active individual energy characteristic, also included the interaction energy characteristic, this two aspects constituted the abnormal behavior's globe energy model feather. It based the calculation method of energy characteristics on the optical flow characteristic, adding the angle variation characteristics to better reflect the behavior of the violence. In the end, it tested the method on diffe-rent types of behavior videos, also compared with the traditional optical flow energy method. The experimental results show the globe energy model is more efficient and accurate, and can achieve the expected purpose.