A behavior recognition method based on self-attention mechanism

Abstract

The invention discloses a behavior recognition method based on a self-attention mechanism. The method adopts a key frame target position prediction and continuous frame action category prediction module based on a multi-angle attention mechanism, and can realize target positioning while completing continuous frame action detection. Function. In the method, the key frame target position prediction based on the multi-angle attention mechanism and the continuous frame action category prediction module are used to replace the 3D convolutional network, which solves the problem of a large amount of calculation for the 3D convolutional network model, and improves the parallelism of the model on the GPU. At the same time, the key frame target position prediction and continuous frame action category prediction module based on the multi-angle attention mechanism can avoid the problem of weak compatibility during model conversion or deployment due to 3D convolution under different deep learning frameworks.

Description

technical field [0001] The invention relates to a behavior recognition method based on a self-attention mechanism, which belongs to the field of human motion recognition. Background technique [0002] Action recognition realizes the action classification task by extracting the action features of continuous video frames, and avoids the occurrence of possible dangerous behaviors in practice, and has a wide range of practical application scenarios. [0003] The existing action recognition methods are all based on 3D convolution, which is used to extract continuous frame features in time series, improve the algorithm's ability to extract image features in time series, and increase the accuracy of action recognition. Compared with 2D convolution, 3D convolution needs to consider time axis features, resulting in a large increase in kernel parameters. At the same time, 3D, as a new computing method, has poor support for 3D convolution under different deep learning frameworks, whic...

AI Technical Summary

Problems solved by technology

Compared with 2D convolution, 3D convolution needs to consider time axis features, resulting in a large increase in kernel parameters

At the same time, as a new computing method, 3D has poor support for 3D convolution under different deep learning frameworks, which affects the practical applicability of action recognition algorithms based on 3D convolution.

Images