【摘要】 异常行为检测在智能监控系统领域中有广泛的应用前景。本文针对此应用领域,提出了一种结合光流特征和梯度直方图特征的视频异常行为检测及定位方法。首先利用视频背景提取算法进行前景提取和标注,实现对前景信息的分割。然后利用光流和梯度直方图特征提取算法对前景图像分别提取光流和梯度直方图特征,其次,使用支持向量机对数据进行训练和测试。最后结合光流幅度信息与前景标记信息对判断出来的异常行为进行定位。实验结果表明,与先前算法相比,本文算法可以检测出异常行为,并且能够对异常帧进行异常行为定位。更多还原

【Abstract】 Abnormal behavior detection has a wild range of applications in the field of intelligent monitoring systems. A video abnormal behavior detection and localization algorithm is proposed that combines the feature of optical flow and histogram of oriented gradients. Firstly,video background extraction algorithm is used to extract and label the foregrounds which could achieve the prospect segmentation information. Then histogram of oriented gradients and optical flow feature extraction algorithm are selected to extract the foreground images’ histogram and optical flow characteristics,following by support vector machines for the training and testing data. Finally,the abnormal behavior is located with information of the optical flow’s magnitude and the foreground label.更多还原