Human Action Recognition Method Based on Recurrent Convolutional Neural Network

Abstract

The invention discloses a human action recognition method based on a cyclic convolutional neural network, which belongs to the fields of image classification, pattern recognition and machine learning. Action recognition accuracy is low. The invention constructs a data set, namely randomly selects sequence pairs with the same length from a public data set, and each frame in each sequence includes an RGB image and an optical flow image; constructs a twin network, and each network in the twin network sequentially includes a CNN layer, an RNN layer and an optical flow image. Temporal Pooling layer; constructs a "recognition-verification" joint loss function; builds a deep convolutional neural network based on data set training and a "recognition-verification" joint loss function; based on the human action sequence pairs to be recognized, trained in turn A deep convolutional neural network and a trained "recognition-verification" joint loss function are used to obtain action class recognition results for sequence pairs. The present invention is used for human action recognition in images.

Description

technical field [0001] A human action recognition method based on a cyclic convolutional neural network is used for human action recognition in images, and belongs to the fields of image classification, pattern recognition and machine learning. Background technique [0002] Human action recognition is one of the hot and cutting-edge research topics in the field of computer vision and machine learning, and has broad application prospects in intelligent video surveillance, intelligent human-computer interaction, and content-based video analysis. [0003] The main problem to be solved in video-based human action recognition is to process and analyze the original image or image sequence data collected by the sensor (camera) through the computer, and learn and understand the human action and behavior. Human action recognition mainly includes the following three steps: first, detect appearance and motion information from image frames and extract low-level features; then model beha...

AI Technical Summary

Problems solved by technology

[0004] In view of the problems of the above research, the object of the present invention is to provide a human action recognition method based on a circular convolutional neural network, which solves the problems in the prior art due to the changes within and between action categories or the video is composed of continuous frames. Causes problems such as low accuracy of human motion recognition

Images