Research On Fatigue Driving Detection System Based On Deep Learning And Multi-attribute Fusion

Fatigue driving always endangers road traffic safety and is one of the main causes of traffic accidents.How to establish a set of effective fatigue condition monitoring and judgment mechanism to achieve fatigue early warning has become a hot spot of scientific research.Traditional detection methods are mostly based on some physiological parameters of human body.Through monitoring the changes of physiological parameters,the fatigue state can be determined.With the rapid development of deep learning in many fields,the fatigue detection method based on computer vision has been widely concerned and gradually moved to the commercial application stage.Compared with the traditional detection method,the fatigue detection method based on computer vision is accurate and reliable,the detection process does not rely on a large number of wearable equipment,and the human-computer interaction experience is good.However,in the specific application scenarios of the algorithm,there are often interference of environmental factors such as partial occlusion of the tested image or low resolution of the image,complex care changes of the detection environment,and instability of the acquisition equipment affected by driving.In the face of many challenges,how to make the detection accuracy and reliability of the fatigue state,at the same time,the algorithm has practical application space,which still needs further exploration.The traditional fatigue detection system based on facial information often detects and analyzes a specific facial attribute.Due to the detection of individual differences and fuzzy definition of fatigue state,the algorithm has the problems of high recognition error rate and poor robustness.Moreover,the single attribute model cuts off the potential mutually exclusive relationship between facial attributes.At the same time,complex models and slower computation speed also limit the application scenarios.In response to these urgent problems,this paper first proposes a fusion model of face detection and pose analysis based on the MTCNN network.At the same time,a neural network forward calculation process simultaneously obtains the position of the face area and the face pose in the image,which improves the completion of the complete face pose analysis task speed.Secondly,based on Squeeze-Net,a network cascade strategy was used to design a facial landmarks detection network to obtain 72 landmarks coordinates of the face.Finally,fatigue characteristics were extracted from the obtained facial attribute information.Decision-making methods.The following innovations were carried out:(1)Face attribute analysis,especially multi-attribute analysis,mostly relies on face detection and face landmarks detection,which lengthens the process of face attribute analysis.In order to solve the above problems,a fusion model of face detection and pose analysis is proposed.The multi task strategy shortens the process of facial pose analysis by integrating facial region detection and facial pose analysis into the same network.This chapter introduces the network structure and parameter definition of each layer of face detection and head posture analysis fusion network in detail,explains the input and output of the network,optimizes the activation function and training strategy used in the training process,and improves the speed and accuracy of detection and recognition while optimizing the model structure based on the actual application scenario of the algorithm.(2)From the perspective of engineering,a cascaded face landmarks detection network based on Squeeze-Net was proposed,and the facial attributes were further analyzed.First,the basic Squeeze-Net network structure and detailed parameters of the fire module are introduced,and then a cascade network structure for eye and mouth landmarks detection is proposed.Use the detection network to locate the 72 landmarks on the face,and select the appropriate analysis strategy for the detected landmarks of the eyes and mouth,and analyze the state of the organs with eyes open,closed,mouth opened,and closed mouth,so as to prevent fatigue in the back.The detection system lays the foundation.(3)In order to solve the problems of low reliability,high false detection rate,incomplete and rich fatigue indicators in the fatigue judgment standard,and easy to be interfered by complex environment detection,a multi feature fusion decision-making method based on information fusion strategy is proposed in this paper.This method uses face region detection results,head posture information and face landmarks information to extract and analyze fatigue features of multiple faces,and uses information fusion strategy to extract fatigue features,so as to improve the recognition accuracy of the algorithm.The real scene data acquisition platform is built,and the judgment strategies after fusion are compared and analyzed.Based on the research results,an online fatigue driving detection system for road traffic scene is built to monitor and warn the fatigue driving of vehicle drivers in real time.Based on the existing research results,the proposed model involves face region detection,face landmarks detection,face multi-attribute analysis and application.Based on the engineering point of view,the model revolves around the lightweight design idea,improves the detection accuracy and running speed,and has the ideal model size.The model has good robustness in the specific application scene.Using the information fusion strategy to propose the multi-attribute fusion fatigue decision strategy,the designed fatigue early warning system has good reliability.The engineering application of facial multi-attribute analysis algorithm from this paper is further explored for others.