A method for detecting abnormal behavior in video

Abstract

The invention discloses a method for detecting abnormal behavior in a video, which includes: extracting all target candidate frames from the video; calculating the optical flow field of each frame image in the video, thereby extracting the motion features in each target candidate frame; calculating the video The convolution features of each frame of image in the image, so as to extract the appearance features of each target candidate frame; all the extracted motion features and appearance features are trained separately to establish a Gaussian mixture model of normal behavior, and the motion of each target candidate frame is tested. The features and appearance features are predicted by the corresponding GMM to obtain the normal score, and then the two normal scores are fused to obtain the minimum value. If the minimum value is lower than the set abnormal threshold, the target individual in the corresponding target candidate box is judged as abnormal . Considering the speed anomaly and appearance anomaly as well as the effectiveness of feature extraction, this method can detect most of the abnormal behaviors in the surveillance video, and has the advantages of good robustness and accurate target location.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a method for detecting abnormal behavior in videos. Background technique [0002] Abnormal behavior detection in densely populated scenes is an important problem in the field of computer vision, with various challenges in video surveillance. The scene dependence of the anomaly detection problem itself determines the complexity of its detection. Different from traditional object detection, anomaly detection first needs to determine the definition of anomaly. Intuitively speaking, anomalies refer to phenomena that do not conform to the constraints of normal behavior or phenomena with relatively low probability of occurrence. The video abnormal behavior detection method automatically detects individuals who do not conform to the normal sample model by capturing the video information of the target area. According to the scope and impact of anomalies, anomalies are divided ...

AI Technical Summary

Problems solved by technology

[0010] However, the current abnormal video behavior detection technology has poor robustness, and there is a problem that it cannot be accurately located.

Images