A kind of abnormal behavior detection method and system

Abstract

The invention discloses an abnormal behavior detection method and system. On the basis of using deep learning technology, the edge server and cloud architecture advantages are integrated to ensure high response and low delay of the edge server. The edge server uploads abnormal behaviors in open scenarios to the cloud and downloads new behavior categories in the cloud system. The cloud uses active label learning to represent abnormal behaviors by known behaviors. As a result, a complete closed loop of abnormal behavior detection between the edge server and the cloud is constructed. The edge server does not need to maintain a long-term connection with the cloud, which reduces the pressure on network transmission. The support for updating the behavior category of abnormal behavior is also more in line with the needs of abnormal behavior detection in open scenarios.

Description

technical field [0001] The invention relates to the field of deep learning, in particular to an abnormal behavior detection method and system. Background technique [0002] The rapid development of smart cities and the large-scale deployment of surveillance equipment and 5G high-speed networks have resulted in a large amount of security video data growing exponentially. What's more, such surveillance video data also contains abnormal behavior data, which seriously endangers the safety of cities. Therefore, how to effectively manage, analyze and mine abnormal behaviors in public places has become one of the most concerned issues in the industry [1] . In the past, the video surveillance system relied on a large number of manual participation, and due to the fatigue of the staff, it was easy to cause missed detection and false detection, and could not guarantee the real-time dynamic analysis of abnormal behavior detection, making the automatic analysis of surveillance video co...

AI Technical Summary

Problems solved by technology

Moreover, the supervised learning process of this type of system inputs a complete video of a complete behavior category, but because certain behaviors often appear repeatedly in the video, the coarse-grained input leads to the final prediction of the behavior category is still not accurate enough [8] , the recognition results can only roughly describe the abnormal type, and cannot show the reasons for the abnormal behavior, including the time and type of behavior, etc. [9][10]

[0005] Specifically, the abnormal behavior detection system [11] Although it is an advanced technology combining the edge server and the cloud, the system has the following technical defects: 1) The predicted behavior category and cycle consistency points of the input video sequence obtained by the system are not used as the input of the convolutional layer, resulting in low accuracy of abnormal behavior recognition; 2) During the operation of the system, it completely relies on the automatic adjustment inside the system, and no external manual intervention parameters are added, which further leads to a low accuracy rate of abnormal behavior recognition; 3) What the system cloud finally provides to the edge server is active learning and training. Abnormal behavior detection model, the edge server needs to replace the previous abnormal behavior detection model, poor operation convenience

[0006] To sum up, on the one hand, the existing abnormal behavior detection cannot meet the needs of abnormal behavior detection in open scenarios due to network transmission delay and supervised learning based on a preset number of behavior category data.

On the other hand, the abnormal behavior detection system [11] Although the edge server, cloud and active learning process are used at the same time, due to a large number of technical defects, it is not effective in identifying abnormal behaviors and operating convenience in open scenarios.

Images