Hybrid neural network complex behavior recognition method based on key frame space-time dimension reduction

By extracting keyframes from human behavior videos and constructing a behavior knowledge graph, and combining hybrid neural networks and dynamic time warping algorithms, the problems of spatiotemporal redundancy, multi-scale fusion, and temporal structure parsing in complex behavior recognition are solved, achieving efficient and accurate complex behavior recognition.

CN122435682APending Publication Date: 2026-07-21JIANGXI UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
JIANGXI UNIV OF TECH
Filing Date
2026-04-29
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing technologies for complex behavior recognition suffer from problems such as spatiotemporal redundancy and computational efficiency, difficulties in multi-scale spatiotemporal feature fusion, insufficient spatiotemporal joint modeling, and difficulties in parsing the temporal structure of complex behaviors.

Method used

By extracting keyframes from human behavior videos, a behavior knowledge graph is constructed and multiple time-length versions are generated. Hybrid neural networks (3D-CNN and Bi-LSTM) are used for modeling, and feature fusion is performed by combining a dynamic gating fusion module and evidence theory. Finally, complex behaviors are identified through a dynamic time warping algorithm.

Benefits of technology

It improves the efficiency and robustness of complex behavior recognition, enhances the ability to perceive changes in behavior rhythm and stages, and can accurately identify complex behavior combinations and locate their occurrence. It is applicable to fields such as intelligent monitoring, human-computer interaction, and medical rehabilitation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122435682A_ABST
    Figure CN122435682A_ABST
Patent Text Reader

Abstract

The application provides a hybrid neural network complex behavior recognition method based on key frame space-time dimension reduction, extracts key frames from a video sequence according to human body behavior posture kinetic energy, constructs a behavior knowledge graph representing human body behavior space-time information by using the extracted key frames, so as to realize space-time dimension reduction of input data, and then uses the behavior knowledge graph for subsequent behavior recognition and positioning operation, so as to obtain the occurrence position of complex behavior in the video. The application can greatly reduce resource consumption by performing space-time dimension reduction on input data.
Need to check novelty before this filing date? Find Prior Art