Cross-domain driving abnormal behavior detection model establishment method and application thereof

By introducing an attention-based incentive network and a domain discriminator, the cross-domain driving abnormal behavior detection model can identify and exchange irrelevant features in cross-domain scenarios, solving the problems of insufficient data annotation and scenario differences, and improving detection accuracy and generalization ability.

CN118230299BActive Publication Date: 2026-07-03HUAZHONG UNIV OF SCI & TECH
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HUAZHONG UNIV OF SCI & TECH
Filing Date
2024-04-11
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing abnormal driving behavior detection systems suffer from reduced detection accuracy and usability in cross-domain scenarios due to insufficient data annotation and scenario differences. Traditional domain adaptation methods fail to effectively extract the importance of sensors and time points, affecting the performance of the detection model.

Method used

A cross-domain abnormal driving behavior detection model is adopted. By introducing an attention-based incentive network and a domain discriminator, semantically irrelevant features are identified and exchanged. A cross-domain feature extractor and classifier are established to enhance feature representation capabilities and improve the model's cross-domain adaptability.

Benefits of technology

It effectively expands the scale and diversity of training data, improves the model's generalization ability to unknown data, increases the accuracy of cross-domain driving abnormal behavior detection, and reduces the dependence on massive labeled data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118230299B_ABST
    Figure CN118230299B_ABST
Patent Text Reader

Abstract

The application discloses a cross-domain driving abnormal behavior detection model establishing method and application thereof, and belongs to the field of driving behavior detection, which comprises the following steps: constructing a source domain data set with driving behavior category labels and a target domain data set without driving behavior category labels, and training a detection network; the detection network comprises: a feature extractor for extracting original features of time-series sensing data; a classifier for detecting whether driving abnormal behavior exists in a corresponding time window of the original features; an excitation network for identifying and exchanging semantic-irrelevant feature parts in initial features of the source domain and the target domain based on an attention mechanism, so as to obtain mixed features of the source domain and the target domain; and a domain discriminator for predicting the domains to which the original features and the mixed features belong; a training loss comprises a classification loss and a domain discrimination loss; and after the training is completed, a network composed of the feature extractor and the classifier is output as a cross-domain driving abnormal behavior detection model. The application can improve the precision of cross-domain driving abnormal behavior detection.
Need to check novelty before this filing date? Find Prior Art