A Human Behavior Recognition Method Combining 3D Jump Connections and Recurrent Neural Networks

Abstract

The invention discloses a human body behavior recognition method combined with a 3D jump connection and a cyclic neural network, comprising the following steps: step 1, dividing each segment of video into N parts, and extracting L frames of pictures from each part, where N and L are natural numbers ; Step 2, use the trained 3D convolutional neural network to extract spatio-temporal features of the video, and connect the spatio-temporal features of different levels in series to obtain high-dimensional feature vectors; Step 3, normalize the high-dimensional feature vectors obtained in step 2 Processing; step 4, sending the high-dimensional feature vector after normalization processing in step 3 to the recurrent neural network for feature fusion; step 5, classifying the fused features in step 4 to obtain the corresponding action category of the video. This method does not need to manually extract low-level motion information. Compared with the artificial motion feature design method, the present invention has better robustness and can effectively process video information for a longer period of time.

Description

technical field [0001] The invention belongs to the technical field of computer vision recognition, and in particular relates to a human behavior recognition method combined with 3D convolution layer jump connections and cyclic neural networks. Background technique [0002] Since human action recognition has important application prospects and market value in the fields of video surveillance, human-computer interaction, and virtual reality, video-based human action recognition has become one of the research hotspots in computer vision. At the same time, as deep learning, especially convolutional neural networks, has achieved effective results in computer vision, human behavior based on convolutional neural networks has attracted the attention of a large number of researchers. [0003] Patent No. CN201611117772.9 "Behavior Recognition Method Based on Trajectory and Convolutional Neural Network Feature Extraction" first extracts trajectory from the input image and video data, ...

AI Technical Summary

Problems solved by technology

In this method, when a video with a long duration is input, the fused feature dimension will be too high, which makes the network more difficult to train

[0007] In summary, although there are many studies on motion recognition based on convolutional neural networks at home and abroad, there are problems such as the need to manually extract motion information from video data in advance or the inability to process long-duration videos.

Images