CAJ | 학술논문

提出了一种基于单目摄像机的异常行为检测方法，可有效检测群聚群散事件。针对传统稠密光流法耗时大，抗噪声性差，对光照变化敏感，而稀疏光流又不能提供丰富运动信息的不足，提出采用二值前景为掩码计算局部稠密光流，既剔除了干扰，又减少耗时。针对光流仅能反应物体瞬时运动信息，不能反应运动的累积效应，不能描述物体间的相互影响，提出建立以光流为底层特征的粒子流场，以稳定地进行特征提取。最后，通过计算粒子幅值直方图的能量来进行异常行为的检测。搭建了实验监控系统，不同场景的实验表明，该方法实时性好，鲁棒。
제출료일충기우단목섭상궤적이상행위검측방법，가유효검측군취군산사건。침대전통주밀광류법모시대，항조성성차，대광조변화민감，이희소광류우불능제공봉부운동신식적불족，제출채용이치전경위엄마계산국부주밀광류，기척제료간우，우감소모시。침대광류부능반응물체순시운동신식，불능반응운동적루적효응，불능묘술물체간적상호영향，제출건립이광류위저층특정적입자류장，이은정지진행특정제취。최후，통과계산입자폭치직방도적능량래진행이상행위적검측。탑건료실험감공계통，불동장경적실험표명，해방법실시성호，로봉。
This paper proposes a novel method to detect abnormal behavior with single camera,which can capture the dynamics of the crowd behavior successfully. Due to the time-consuming,the bad anti-noise and high sensitivity to light for dense optical flow method and inadequate motion information for sparse optical flow,binary foreground mask is used for calculating local dense optical flow. Particle flow is constructed based on optical flow for stable fea-ture extraction,since optical flow only reflects instantaneous motion information and cannot describe the motion ac-cumulation and interaction between objects. A kinetic energy of particle magnitude histogram is computed to capture the abnormality. A video surveillance system is constructed,the experiments on different scenes show that the pro-posed method is real-time,and robust.