Action recognition method and device and storage medium

Abstract

The invention provides an action recognition method and device and a storage medium, and the method comprises the steps: carrying out the processing of obtained to-be-detected video data, and obtaining a plurality of feature matrixes; the plurality of feature matrixes are obtained by processing a plurality of video frames containing the target object; processing the plurality of feature matrixes by using a time sequence feature extraction network model to obtain a plurality of time sequence action feature matrixes corresponding to the plurality of feature matrixes; wherein a network layer in each network block in the time sequence feature extraction network model carries out fusion reuse and grouping processing on a preorder feature matrix; the preorder feature matrix comprises a feature matrix input by a network block corresponding to a network layer and a feature matrix output by a preorder network layer ranked in front of the network layer; and determining action recognition results of the target object in the plurality of video frames through the plurality of time sequence action feature matrixes. According to the scheme, the occupation of computing resources can be reduced, and the recognition precision is also improved.

Description

technical field [0001] Embodiments of the present invention relate to the technical fields of computers and the Internet, and in particular, to a method, an apparatus, and a storage medium for motion recognition. Background technique [0002] The traditional action recognition method uses the VGG (Visual Geometry Group Network) model, and uses several consecutive convolution kernels to replace the larger convolution kernel for action recognition. The model increases the depth of the network under the premise of having the same perceptual field of view, which proves that increasing the depth of the network can affect the performance of the network to a certain extent and improve the effect of the network, but the model uses more parameters and consumes more computing resources. Another action recognition method uses the GoogLeNet model to turn the full connection into a sparse connection. While increasing the depth and width of the network, the parameters of the network are ...

AI Technical Summary

Problems solved by technology

This model increases the depth of the network under the premise of the same perceptual field of view, which proves that increasing the network depth can affect the performance of the network to a certain extent and improve the effect of the network, but the model uses more parameters and consumes more computing resources

Another action recognition method uses the GoogLeNet model to change the full connection into a sparse connection. While increasing the depth and width of the network, it reduces the parameters of the network to a certain extent, but as the number of network layers increases, the network will degenerate. , leading to inaccurate identification of

Therefore, there are problems in the related technologies, such as taking up more computing resources and poor recognition accuracy.

Images