A security abnormal behavior recognition method and system based on multi-modal perception

CN122435289APending Publication Date: 2026-07-21WUHAN HONGZHANBO INTELLIGENT TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
WUHAN HONGZHANBO INTELLIGENT TECH CO LTD
Filing Date
2026-05-08
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Traditional multimodal fusion methods cannot handle the differences in visible light and thermal infrared modal pointing caused by low-emissivity glass or curtain wall interfaces, leading to misjudgments or missed detections of abnormal behavior and affecting the reliability of security monitoring.

Method used

By extracting the edge contour lines of low-emissivity glass or curtain wall interfaces, visible light and thermal infrared video sequences are mapped to a local coordinate system. The edge motion density and vertical compression of thermal radiation response are calculated to construct an edge dynamic map. The optimal behavior path is obtained using an ant colony search algorithm to identify abnormal behavior.

Benefits of technology

It effectively overcomes the modality separation problem, improves the accuracy and robustness of abnormal behavior recognition, and significantly enhances the ability to recognize boundary-attached and boundary-separated abnormal behaviors in complex security scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122435289A_ABST
    Figure CN122435289A_ABST
Patent Text Reader

Abstract

The application relates to the security technology field and discloses a security abnormal behavior recognition method and system based on multi-modal perception, which comprises the following steps: extracting the edge contour line of an interface, mapping visible light and thermal infrared video sequences into a local coordinate system with an arc length direction and a vertical direction; obtaining an edge visible light motion density, and extracting an edge thermal radiation response and a vertical compression degree; combining the above parameters to calculate time lag strength, optimal delay amount and parallel movement strength, and constructing an edge dynamic graph; extracting local extreme points on the edge dynamic graph as search nodes, performing ant colony search to obtain an optimal behavior path; and finally positioning an abnormal period and area according to the path, and converting the abnormal period and area into a behavior description vector to output an abnormal category and strength. The application effectively solves the cross-modal heterogeneous interference and object separation problem, and improves the recognition precision of boundary abnormal behavior.
Need to check novelty before this filing date? Find Prior Art