Abnormal Event Detection in Video Based on Sparse Representation

As a research hotspot in intelligent video surveillance system, abnormal event detection has attracted the attention of many researchers in recent years. In order to overcome the shortcoming of the semi-supervised model, that is, the training sample is difficult to contain all possible situations, which leads to the occurrence of error detection, we propose a method based on sparse representation. The principle of this method is to train the model with normal data and abnormal data respectively, get two sparse representation models, and then judge whether there are abnormal events according to the results of the two models. The method has passed the test of the existing data sets and achieved good results.