A behavior recognition method for intelligent video analysis

By collecting video frame sequences in intelligent video analysis for human detection and cross-frame tracking, generating short-term and long-term anchor points, and combining background structure degradation index and trajectory semantic breakage index for drift situation discrimination, the problem of instability in existing behavior recognition systems is solved, and stable and continuous output of behavior recognition is achieved.

CN121884463BActive Publication Date: 2026-05-29XIAN XINGXUN INTELLIGENT COMM TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
XIAN XINGXUN INTELLIGENT COMM TECH CO LTD
Filing Date
2026-03-20
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

In existing intelligent video analytics, behavior recognition systems are prone to uncertainty fluctuations when faced with chronic changes such as lighting variations, lens contamination, and background texture changes, leading to unstable behavior labels and affecting the effectiveness of practical applications.

Method used

Human detection and cross-frame tracking are performed by acquiring video frame sequences, generating target trajectory sequences and extracting behavioral feature sequences. Short-term and long-term anchor points are established, and drift status is determined by combining background structure degradation index and trajectory semantic breakage index. Feature normalization and anchor point reconstruction are performed to achieve stability and continuity of behavioral output.

Benefits of technology

Under varying lighting and image clarity, the behavior recognition system can maintain category stability, reduce category drift and confidence ranking fluctuations caused by slow scene changes, and improve the long-term stability and continuity of behavior recognition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121884463B_ABST
    Figure CN121884463B_ABST
Patent Text Reader

Abstract

The application discloses a behavior recognition method for intelligent video analysis, and particularly relates to the field of computer vision, and aims to solve the technical problems that the behavior output is unstable and the calibration is easy to spread due to the superposition of imaging slow variation and occlusion jitter in a monitoring scene; a target trajectory sequence is obtained by collecting a video frame sequence and completing human body detection and cross-frame tracking, a short-time behavior representation is generated by extracting a behavior feature sequence along the target trajectory, and a short-term anchor point and a long-term anchor point are established; a background structure degradation index and a trajectory semantic fracture index are calculated in a short-time segment to generate a drift state judgment result, and a domain state is determined in combination with a trend continuity rule; feature normalization mapping or anchor point reconstruction is performed according to the domain state to form original candidate behavior results and calibration candidate behavior results, consistency verification is performed to determine a final behavior output, and the short-term anchor point and the long-term anchor point are synchronously revised, so that long-time stable recognition and continuous output are realized.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • CN115035463A

  • CN120823548A

  • CN121074785A