Abstract:

At present， most anomaly detection algorithms only train models through normal samples， but the lack of abnormal sample may cause misjudgment. This paper proposes an anomaly detection algorithm based on effective anomaly sample construction. The clustering clusters representing different types of normal events are obtained by K-means clustering algorithm. Then， the abnormal samples are constructed based on the temporal relationship of abnormal events. Combined with the abnormal samples constructed in this paper， the classifier is trained by binary classification support vector machine algorithm to transform the detection task into classification task， thereby improve the detection accuracy. The effectiveness of the algorithm is verified on the classic dataset（the Avenue dataset）， and it is found that the detection accuracy of this algorithm is better than some advanced algorithms. Therefore， making full use of the timing relationship of video to construct abnormal samples can effectively improve the effectiveness of abnormal detection.

Key words: communication and information system, video processing, abnormality detection, support vector machines