Video streaming anomalous event detecting method based on measure query entropy

Abstract

The invention discloses a video streaming anomalous event detecting method based on measure query entropy, and belongs to the technical field of digital image processing. A weak supervision joint subject model is used as an initial model. A data stream including an unmarked normal class and an anomalous class is given, and whether a sample needs to be marked or not at the current moment and which is used for marking the sample are judged in sequence. The parameters of the current model are updated continuously, and meanwhile a threshold value is updated. The model parameters and the threshold value are updated repeatedly until traversal is carried out on the whole data stream. In the testing process, a likelihood threshold value of a test data set and the model obtained at last is calculated, and therefore an anomalous event can be recognized. A time cause and effect guide model is introduced to an active learning query strategy to collect more anomalous event samples. Two query criteria are adopted at the same time, an unknown sample, an uncertainty sample and the recognized anomalous sample are utilized to update the model, the new measure query entropy is designed, the classification accuracy is combined, and the performance of the model is monitored in real time.

Description

technical field [0001] The invention relates to a method in the technical field of digital image processing, in particular to a video stream abnormal event detection method based on measurement query entropy. Background technique [0002] In video surveillance, the identification of abnormal events is an important task and has received much attention. Nevertheless, it is still difficult and faces many challenges in practical environment. First, abnormal events are often unpredictable. Second, both normal events and abnormal events are inherently diverse. Another most critical problem is the lack of enough labeled samples for model training and verification, which is especially prominent for abnormal events. Furthermore, modeling anomalous events is not trivial even given training samples. In the big picture, in large complex scenes, anomalous events often seem small compared to the normal events that are going on. [0003] After searching the existing technical literatu...

AI Technical Summary

Problems solved by technology

The invention solves the problems of abnormal event modeling and abnormal event sample collection, and ensures the robustness and effectiveness of abnormal event identification

Images