A transformer-based behavior analysis method, device and terminal equipment thereof

Abstract

The invention discloses a transformer-based behavior analysis method, device and terminal equipment thereof. The method includes: step 1, splitting a human body image into multiple limb parts through a preprocessing module, and step 2, expanding each sub-block into a Dimension vector, as the input of the transformer model, the transformer model includes an encoder and a decoder; step 3, optimize the sub-block through the polymorphic fusion module and enhance the global information at the same time to obtain the behavior category label. The present invention makes the link between target local and local, local and global closer, enables the model to focus on subtle actions, greatly improves the accuracy of behavior recognition, and thus achieves the purpose of precise early warning of dangerous behavior.

Description

technical field [0001] The present invention relates to the field of image recognition technology research, especially neural network, and in particular to a transformer-based behavior analysis method, device and terminal equipment. Background technique [0002] With the vigorous promotion of smart cities, the smart social governance system is gradually improving, and the mediation of social conflicts, as an important part of social governance, has also attracted much attention. In conflict mediation, due to the different education levels and personal qualities of the parties themselves, their performance in the mediation process is also different. Some parties are too excited and may have physical conflicts. This not only hinders the normal work order, but also affects The legitimate rights and interests of other masses have greatly affected the construction and development of a harmonious society. Therefore, it is particularly important to analyze the behavior of the part...

AI Technical Summary

Problems solved by technology

[0003] In the process of realizing the present invention, the inventors found that there are at least the following problems in the prior art: behavior recognition technology includes data analysis and image recognition, data analysis relies on sensing instruments to collect multiple specific indicators of the target, and comprehensively outputs results based on numerical statistics. This type of method is time-consuming and labor-intensive, and has limited accuracy. Most of the existing image recognition methods are based on deep learning technology, using a single convolutional neural network to extract behavioral features, only considering global features, and insufficient expression of local features.

Some deep learning methods take into account local information and incorporate attention mechanisms, but the fusion method of local features and global features is relatively simple, the feature fusion is not comprehensive enough, and the expression ability is not strong enough

Images