Driver Distraction Detection With Hybrid Transformer-CNN Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current driver distraction detection methods, primarily based on convolutional neural networks (CNN), face challenges such as manual parameter setting, limited global feature capture, and increased computational complexity, which hinder effective detection of diverse driver distractions in real driving scenarios.
Innovation Solution
A transformer-based driver distraction detection method integrating a Twins transformer with CNN, incorporating a residual embedding module, group convolution, and label-smooth cross-entropy loss, to enhance feature extraction and classification accuracy, forming the driver distraction detection model FPT.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If CNN is used for feature extraction, then local features can be captured, but global feature capture capability is limited
Solution Approach 1:
The image is divided into multiple patches that are processed by the transformer encoder, allowing both local patch-level features and global image-level features to be captured through the self-attention mechanism across all patches
Solution Approach 2:
The patent combines CNN and transformer architectures into a hybrid model where CNN extracts local features from image patches and the transformer captures global dependencies, merging the strengths of both approaches to achieve comprehensive feature extraction
2Measurement precision
If network depth is increased to improve feature extraction, then classification accuracy improves, but computational complexity increases
Solution Approach 1:
The transformer encoder dynamically adjusts attention computations based on input characteristics, and the model uses adaptive feature fusion mechanisms that optimize computational resources while maintaining high classification accuracy across different network depths
Solution Approach 2:
The patent employs adaptive parameter adjustment where the model dynamically modifies its internal parameters and feature representation dimensions based on input complexity, reducing unnecessary computations while preserving accuracy
3Ease of manufacture
If manual parameter setting is used in CNN, then model flexibility is reduced, but implementation simplicity is maintained
Solution Approach 1:
The transformer-based model performs automatic feature extraction and representation learning without requiring manual parameter tuning, with the self-attention mechanism adaptively learning optimal feature weights and relationships from the data itself
Solution Approach 2:
The patent uses learnable parameters in the transformer encoder that are automatically optimized during training, replacing fixed manual parameters with adaptive parameters that change based on input characteristics and learning objectives
Data Source
AI summary
The present disclosure provides a transformer-based driver distraction detection method and apparatus, belonging to the field of driving behavior analysis. The method includes: acquiring districted driving image data; building a driver distraction detection model FPT; inputting the acquired distracted driving image data into the driver distraction detection model FPT, analyzing the distracted driving image data by using the driver distraction detection model FPT, and determining a driver distraction state according to an analysis result. The present disclosure proposes a new network model, i.e., a driver distraction detection model FPT, based on Swin, Twins, and other models. Compared with a deep learning model, the FPT model compensates for the drawback that the deep learning model can only extract local features; the FPT model improves the classification accuracy and reduces the parameter quantity and calculation amount compared with the transformer model. The present disclosure adjusts the loss function of the whole network and adds label smoothing to the cross-entropy loss function, to increase the accuracy of classification, effectively suppress overfitting, and improve the detection accuracy.