Double-person interactive behavior identification method in complex background

Abstract

The invention discloses a double-person interactive behavior identification method in a complex background. The method comprises the following steps of constructing step: a two-layer convolution superposition ISA network used for extracting the spatio-temporal characteristics of videos based on video automatic learning, wherein the construction step specifically comprises the three operations of local spatio-temporal sample extraction, graph layer ISA1 feature extraction and graph layer ISA2 feature extraction; training step: adopting all double-person interactive behavior videos in a trainingset, extracting the spatio-temporal characteristics of the videos in the training set by means of the two-layer convolution superposition ISA network, and obtaining an SPN model structure based on the spatio-temporal characteristics and through the SPN structure learning algorithm; identification step: adopting double-person interactive behavior videos in a test set, extracting the spatio-temporal characteristics of the videos in the test set by means of the two-layer convolution superposition ISA network, and obtaining the identification result of double-person interactive behaviors based onthe SPN model. According to the invention, based on spatio-temporal characteristics extracted through the ISA network and the SPN model structure, a method applied to the universality of double-person behavior videos of different degrees of complexity is provided.

Description

technical field [0001] The invention relates to a two-person interactive behavior recognition method under a complex background, in particular to a double-layer convolution superimposed ISA spatio-temporal feature for video extraction and an SPNs classification structure based on SPN structure learning, and belongs to the technical field of two-person behavior recognition. Background technique [0002] Using information technology to automatically recognize human behavior in video is a hot and key issue in the field of computer vision in recent years. In recent years, using information technology to automatically identify human behavior in real life has become an emerging social demand, and video-based human behavior recognition has practical value. With the development of related technologies in the field of computer vision and the rise of deep learning technology in recent years, a high recognition accuracy rate has been achieved for human behavior recognition in an ideal ...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to solve the problem of two-person interaction in video, such as lens angle conversion, illumination color change, complex scene and noise background, and human body self-occlusion and partial occlusion. The traditional method has great limitations. Technical defects, proposed A video classification method based on ISA spatio-temporal features and SPN with universal applicability to video complexity

Images