基于光流及轨迹的人群异常行为检测

Crowd abnormal behavior detection based on optical flow and track

针对复杂背景及遮挡等原因引起人群异常行为检测性能低的问题,本文提出了一种基于综合光流特征描述符(SOFD)及轨迹的人群异常行为检测方法。首先,根据人群光流场变化计算人群运动速度、加速度、方向和能量,并利用上述特征设计新的时空特征描述符,即SOFD。其次,利用KLT追踪算法获得人群运动轨迹单帧图。最后,基于所获取的上述特征,设计双流卷积神经网络(TS-CNN)以检测人群异常行为。仿真结果表明:与现有主流算法相比,复杂环境下本文方法具有较高的异常行为检测准确率、较好的泛化性及稳健性。

Focusing on the problem of low detection performance of crowd abnormal behavior conducted by complex background and occlusion as well as other factors,a crowd abnormal behavior detection method is proposed based on synthetic optical flow feature descriptor(SOFD)and trajectory.First,the velocity,acceleration,direction and energy of crowd motion are calculated according to the change of crowd optical flow field.Then,a new space-time feature descriptor,i.e.,SOFD,is constructed based on the above characteristics.Third,the Kanade-Lucas-Tomasi(KLT)tracking algorithm is employed to obtain the single frame of crowd motion trajectory.Finally,the two stream convolution neural network(TS-CNN)is depicted with the abovementioned characteristics to detect the crowd abnormal behavior.Compared with the existing state-of-the-arts algorithms,simulation results show that the proposed method has higher accuracy,better robustness and wide application range in abnormal behavior detection under complex environments.