Unsupervised Driver Distraction Detection via RepMLP
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current driver distraction detection methods rely on supervised learning, which requires extensive labeled data, lacks universality, and struggles with global feature capture, leading to limited generalization and high resource consumption.
Innovation Solution
An unsupervised learning method using a RepMLP network with multilayer perceptron (MLP) backbone, residual projection heads, and a contrastive learning-based loss function with stop-gradient strategy for fine-grained driver distraction detection, leveraging unlabeled data and improving global perception.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised convolutional neural network models are used to detect driver distraction, then detection accuracy is improved, but the requirement for labeled data increases significantly
Solution Approach 1:
The system performs self-supervised learning by automatically generating labels from the data itself through contrastive learning. The model learns to distinguish between augmented versions of the same image (positive pairs) and different images (negative pairs), enabling the system to train without external labeled data while maintaining detection accuracy.
Solution Approach 2:
The framework performs preliminary feature extraction and representation learning using unsupervised contrastive learning before actual distraction detection. This preliminary action of learning robust features from unlabeled data prepares the model for accurate detection without requiring extensive labeled training data.
2Device complexity
If manually selected key features are used in traditional machine learning methods, then device complexity is reduced, but adaptability to different driving scenarios deteriorates
Solution Approach 1:
The patent replaces manual mechanical feature selection with automated deep learning feature extraction. The RepMLP network automatically learns relevant features from raw images, substituting the manual process with an intelligent system that adapts to different scenarios without requiring manual reconfiguration.
Solution Approach 2:
The feature extraction process becomes dynamic and adaptive through the unsupervised learning framework. The model automatically adjusts which features are important based on the input data distribution, enabling adaptability across different driving scenarios while maintaining relatively simple implementation.
3Measurement precision
If CNN models with multiple convolutional layers are used, then local feature perception is improved, but global scene perception deteriorates
Solution Approach 1:
The patent transitions from traditional CNN's spatial dimension processing to RepMLP's parameterized transformation approach. This dimensional change allows the model to capture both local and global features simultaneously by operating in a transformed feature space that preserves both fine-grained and holistic information.
Solution Approach 2:
The RepMLP backbone network serves multiple functions: it performs local feature extraction like CNNs while simultaneously capturing global scene context. This universal architecture eliminates the trade-off between local and global perception by integrating both capabilities in a single framework.
Data Source
AI summary
The present disclosure provides a method for fine-grained detection of driver distraction based on unsupervised learning, belonging to the field of driving behavior analysis. The method includes: acquiring distracted driving image data; and inputting the acquired distracted driving image data into an unsupervised learning detection model, analyzing the distracted driving image data by using the unsupervised learning detection model, and determining a driver distraction state according to an analysis result. The unsupervised learning detection model includes a backbone network, projection heads, and a loss function; the backbone network is a RepMLP network structure incorporating a multilayer perceptron (MLP); the projection heads are each an MLP incorporating a residual structure; and the loss function is a loss function based on contrastive learning and a stop-gradient strategy.