Video human body behavior recognition algorithm based on double-flow space-time decomposition

Abstract

The invention provides a video human behavior recognition algorithm based on double-flow space-time decomposition, which improves the input characteristics of a double-flow network, uses an I frame of a compressed video to train a spatial flow network and a P frame to train a time flow network, retains the basic framework of the double-flow network, proposes a new double-flow space-time decomposition convolutional network, and improves the recognition efficiency. A 3D convolutional network in a 3D residual convolutional network (ResNet3D) is split into a hybrid network of a two-dimensional space convolutional network and a one-dimensional time convolutional network, so that a model can effectively acquire time sequence information by using the 3D convolutional network, the parameter quantity of network training is reduced, and the network is easier to optimize.

Description

technical field [0001] The invention relates to the technical field of human action recognition, in particular to a video human action recognition algorithm based on dual-stream spatiotemporal decomposition. Background technique [0002] In recent years, with the substantial increase in computer computing power and the continuous introduction of large-scale data sets, many studies have proved that deep convolutional networks can achieve excellent performance and recognition results in the field of video human behavior recognition, and motion modeling based on deep learning The research focus of the human behavior recognition method is mainly to build a model with excellent discriminative power. At this stage, the mainstream convolutional neural network framework includes long short-term memory network (Long Short-Term Memory Network, LSTM), two-stream network (Two-stream Network ) and 3D convolutional networks, etc. [0003] The dual-stream network is the most representativ...

AI Technical Summary

Problems solved by technology

[0007] However, the traditional 3D convolutional network model has a large number of parameters, and requires high memory and computing power for network training. The existing hardware equipment conditions are difficult to support the development of 3D convolutional network, which will seriously limit the efficiency of the model. How to reduce the 3D It is one of the important research directions of the 3D network framework to increase the parameter quantity of the network and improve its performance of extracting time series information.

Images