Research And Implementation Of Driver Behavior Recognition System In Internet Of Vehicles

Traffic accidents occur frequently in the current society,and a large part of traffic accidents are caused by unsafe behaviors of drivers during driving.In order to reduce the occurrence of traffic accidents,this thesis proposes a set of methods for detecting driver behavior,which can identify the driver’s behaviors that hinder driving safety,such as using mobile phones,smoking,drinking,and scratching the head during driving.The main work of this thesis is as follows:1.Since there is no suitable public dataset for the task of this subject on the Internet,we collected and established the dataset by ourselves.The total size of the dataset reached8,900,and the driver’s location and behavior were labeled.2.A method of locating the driver’s area based on the image is proposed.This method improves the network structure on the basis of the YOLO object detection algorithm,and optimizes the parameters according to the characteristics of the dataset of this subject.Using the data set of this subject for training and testing,the results show that the overall recognition accuracy of the algorithm on the dataset has reached 98%,and it satisfies the real-time requirements well.3.Based on the previous detection of the driver’s area,this thesis proposes a method to recognize the driver’s behavior based on the key feature points of the human body.This method uses the stacked hourglass model to identify the key feature points of the driver’s body,including areas such as the face,hands,and wrists,and then uses the VGG model to extract features from the entire image,and uses the ROIPooling layer to combine the features of the key areas with the entire image.The features of the image are mapped,then the cascade operation is performed and a classifier is used to obtain the final prediction result.The dataset of this subject is used for training and testing the entire algorithm flow.The results show that the overall recognition accuracy of the algorithm on the dataset reaches 92%,and it satisfies the real-time requirements well.In addition,the method is compared with methods directly using convolutional neural networks,and the experiment proves that this method can further improve the recognition accuracy.4.Based on the previous algorithm,this thesis designs a driver behavior recognition system.The client is deployed on the Linux embedded development board,and the server is deployed on the server cluster.The client supports two modes of online detection and offline detection to detect driver behavior.After detecting the driver’s unsafe behavior,the system will record the data in the background and prompt the driver to pay attention to safety,so as to help the driver develop good driving habits and reduce the occurrence of traffic accidents.