Driver Fatigue Recognition Based On Deep Learning

Fatigue driving is very likely to cause major traffic accidents.The detection of driver fatigue has important research significance for preventing traffic accidents.Based on this,relevant scholars have carried out a lot of research in this field and have achieved a series of research results.Among them,the method based on visual appearance features has gradually become the research focus of practical technology in this field with its advantages of no interference to the driver and easy layout of the visual system.However,since the visual image is greatly affected by environmental factors such as external illumination changes and image background,further improvement in system reliability is required.Therefore,based on the analysis of various apparent characteristics reflecting the fatigue state of the driver,the driving fatigue detection model is constructed by using the more robust deep learning algorithm,and the constructed model is verified combined with the experimental data.Aiming at the shortcomings of the above image processing methods,such as low anti-interference and low accuracy,this paper extracts the spatial features of the driver's facial image through the convolutional neural network in deep learning.Considering the dynamic changes of driving fatigue characteristics in time series,this paper combines the temporal expression characteristics of Long short term memory(LSTM)to construct the basic model of fatigue detection.On this basis,in order to make up for the lack of motion characteristics caused by interval sampling,this paper optimizes the basic model by introducing the optical flow characteristics reasonably,and then forms a dual-flow model based on RGB-O image.Finally,based on the timeliness requirements of the system,this paper proposes a dynamic adaptive data sampling mechanism to effectively improve the actual operating efficiency of the system.In this paper,the research on driving fatigue detection focuses on the construction of fatigue detection model and the timeliness of the system.Among them,the key breakthrough technologies include: Firstly,based on the analysis of the dynamic characteristics of the apparent characteristics,this paper makes a comprehensive decision on the identification of driving fatigue from the perspective of time domain and airspace.Secondly,in view of the timeliness requirements of the actual application of the system,this paper effectively improves the operating efficiency of the model through the adaptive dynamic data sampling mechanism.