Abnormal Event Detection in Video Based on SVDD

Abstract

Abnormal event detection, as a hot research field in intelligent video monitoring system, has attracted many researchers' attention in recent years. In order to overcome the shortcomings of the semi-supervised model, namely the training sample is difficult to contain all possible situations, leading to the occurrence of error detection, we propose a method based on support vector data description (SVDD). The principle of the method is to train the model with normal data and abnormal data respectively to obtain two SVDD models, and then judge whether there are abnormal events according to the results of the two models. This method has been tested by existing data sets and achieved good results.