欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
提出了一种基于目标距离关系和图像光流特征的异常行为检测新方法。采用三帧差分法从视频序列中提取出运动前景，利用卡尔曼滤波跟踪运动目标，并采用欧式距离计算目标质心间距离。若距离小于设定阈值，认为可能发生异常行为，采用Lucas—Kanade法提取当前帧的光流信息和光流方向直方图描述行为，计算归一化直方图的熵来判断行为的异常。本方法仅对距离小于阈值的图像帧进行光流计算，弥补了光流法计算量大的缺点，满足了系统实时性需求，基于不同场景的实验结果验证了所提方法的有效性。
英文摘要：
A novel abnormal behavior recognition method is proposed based on object distance rela- tionship and optical flow. The moving foreground region is extracted by three-frame difference method Kalman filter is used to track each moving target so as to get their position information, and obtain dis- tance through calculating euclidean distance between their centroids. If the distance of moving targets is less than the threthold, the moving objects is near enough, an abnormal behavior may happen. The Lu- cas-Kanade algorithm is used to extract the optical flow information of current frame, and the flow features are described by orientation histogram. The behavior is determined through the entropy of normalized ori- entation histogram. This method overcomes disadvantages of large computation of optical flow, and meets the requirements of real-time systems. The experiment results in different environments show the effec- tiveness of the proposed method.
同期刊论文项目
同项目期刊论文
期刊信息
位置：