基于全局-局部选择尺度注意力的异常驾驶状态识别方法

By employing a global-local selective scale attention approach, combined with ResNet50 and Transformer models, global and local features are extracted and fused, addressing the issue of low accuracy in abnormal driving state recognition in existing technologies and achieving more efficient abnormal driving behavior detection.

CN118537844BActive Publication Date: 2026-07-17HEFEI UNIV OF TECH
View PDF 4 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HEFEI UNIV OF TECH
Filing Date
2024-05-31
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing technologies only use global image features to identify abnormal driving states, resulting in low accuracy and difficulty in effectively identifying driver fatigue and distraction.

Method used

A global-local selective scale attention approach is adopted. Convolutional features of video frames are extracted using the ResNet50 model, and global and local attention features are calculated by combining the Transformer model. Multi-scale feature fusion is then performed to train an abnormal driving behavior detection model.

Benefits of technology

It improves the accuracy of identifying abnormal driving conditions, can better utilize image information, focuses on local details of the driver at a specific scale, and identifies distraction and fatigue.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118537844B_ABST
    Figure CN118537844B_ABST
Patent Text Reader

Abstract

本发明公开了一种全局‑局部选择尺度注意力的异常驾驶状态识别方法。本发明使用全局处理时，生成多尺度的全局图像，并估计各尺度全局图像特征的Transformer注意力。融合多尺度Transformer注意力，并使用卷积操作，生成各尺度的选择注意力。将各尺度的选择注意力，与各尺度的全局图像特征融合，获得融合尺度的全局特征。本发明使用局部处理时，首先进行网格划分获得头部局部区域，并生成多尺度局部图像，并计算局部图像特征的各尺度选择注意力，并最终获得融合各尺度的局部特征。本发明关注于多尺度注意力的选择过程，能够注意到驾驶人员特定尺度的局部细节特征，用于识别分心、疲劳的异常状态。
Need to check novelty before this filing date? Find Prior Art