Research On Personalized Health Detection And Navigation For Drivers

According to the World Health Organization,road traffic accidents cause about1.3 million deaths and 50 million injuries worldwide each year.One of the important causes is the cognitive and operational degradation caused by the negative health state of drivers.Therefore,it is urgent to reduce traffic accidents and improve driving safety by detecting drivers’ stress and fatigue in real time from the source of road traffic accidents,building an optimal navigation system for driver health status,and taking corresponding early warning and optimization measures to ensure that drivers are in healthy states and cognitive levels suitable for driving.The development of multimedia technology and sensing technology has made it easy to obtain multimodal data of the driver,vehicle,and environment in real time.However,the multimodal data in driving scenarios with multiple sources of heterogeneity and the difficulty of labeling the datasets pose serious challenges for driver health detection and optimization.Therefore,this thesis is based on attention mechanisms,multimodal fusion methods,and self-supervised learning in order to build multimodal fusion models and self-supervised models for driving scenarios to detect the key factors(stress and fatigue)that affect the driver’s mental and physical health status during driving,so as to accurately identify the driver’s stress and fatigue levels.Then the cybernetic-based driver personalized health navigation method is proposed,and a driver-oriented personalized health navigation system is developed to optimize driver health and enhance driving safety.The main innovation of this thesis is summarized as follows:(1)An attention-based CNN-LSTM multimodal fusion modelIn order to handle multimodal data of drivers with multiple heterogeneous sources,this thesis proposes an attention-based convolutional neural network-long short-term memory network(CNN-LSTM)multimodal fusion model to fuse non-intrusive data from driving scenes,including driver’s eye data,vehicle dynamics data,and environmental data.The multimodal fusion model can automatically extract features from each modality,and use a self-attention layer to assign different weights to features from different modalities for effective fusion to accurately detect driver stress levels.Extensive comparative experiments on a dataset collected by an advanced driving simulator validate the superior performance of the proposed method compared with the state-of-the-art CNN-LSTM model.(2)An isotropic self-supervised learning modelTo address the lack of enough samples in driver fatigue datasets,which makes it difficult for traditional deep learning models to learn effective fatigue state representations from images or videos,this thesis proposes the isotropic self-supervised learning model(Iso SSL-Mo Co)that enables image encoders to learn robust fatigue state representations on publicly available datasets without relying on human-supplied annotations.To enhance the complementarity of multimodal data,an attention-based multimodal fusion model for driver fatigue detection is proposed,and the image encoder is pre-trained with the multimodal fusion model using self-supervised learning.Extensive comparative experiments on the NTHU-DDD and Yaw DD datasets demonstrate the effectiveness of the Iso SSL-Mo Co model and the multimodal fusion model in detecting driver fatigue.(3)A cybernetic-based driver personalized health navigation methodIn order to solve the problems of single driver health detection dimension and lack of personalization of driver health optimization,this thesis firstly proposes a personalized driver health navigation architecture and constructs a personalized comprehensive health index model integrating multiple dimensions such as driver stress and fatigue.Then,a cybernetic-based driver personalized health navigation method is proposed,which sets up a target health state optimization route and builds a closedloop feedback system to continuously improve the driver’s health state.Finally,a demonstration system for personalized driver health state navigation is developed.