Real-time detection method for abnormal behavior

The embodiment of the invention discloses a real-time detection method for an abnormal behavior. The method comprises the following steps: storing a frame of image in a video stream as a still image; subtracting the still image from a current frame in a video stream to perform target detection, performing processing to obtain a binary image, performing connected branch extraction, and finding out a bounding box and an area size of each connected branch; analyzing each connected branch by using the aspect ratio of the bounding box and the area size, and extracting an interested target; using a tracking algorithm to extract and track a behavior track of the interested target; and judging the behavior track. By adopting the method, the normal behavior data sample is used for distinguishing, the abnormal behavior can be automatically identified in real time, the calculation overhead is low, the method can be implemented on small systems such as an embedded system, and the method has a wide application prospect. 本