A Method of Abnormal Behavior Detection for Safety Site Surveillance

Abstract

In order to accurately detect and give alerts to the anomalies in visual images, this paper proposes an image anomaly detection method. For the complex background in the image, a multiframe differential superposition algorithm is proposed to denoise the target image; a feature extraction method is given to extract features for the target image, and then a more complete image with target features is obtained after filtering; a normal behavior model is established to extract the motion information of the target from a single frame of the image; an abnormal detection method is proposed to determine whether it belongs to abnormal behavior. The experimental results show that the accuracy of the abnormal behavior detection method proposed in this paper can better discern the beginning and end of behavior occurrence, abnormal behavior prediction, behavior online detection, and other aspects from the visual image data stream, and the correct detection rate is more than 90%, which reduces the consumption of human resources. At the same time, compared with the existing anomaly detection methods, this anomaly detection presented in this paper not only has higher accuracy, faster speed, and stronger anti-interference ability but also has a better detection effect. These researches advance in this paper can provide a new method and decision support for abnormal behavior detection and identification in a variety of scenarios.