Assisted Lane-Change Technologies Based On Multi-Source Information Fusion For Intelligent And Connected Vehicles

With the rapid development of intelligent and connected vehicle technology,the driving environment of intelligent and connected vehicles has also begun to expand from closed park scenes to open scenes such as urban roads and highways.Lane keeping and changing are the most basic working conditions of intelligent and connected vehicles in these scenarios.How to effectively predict lane-change behavior and provide safety assistance,prevent accidents,and improve the safety of intelligent and connected vehicle has always been a hot issue in the field of intelligent and connected vehicle technology.Researchers have carried out in-depth research on vehicle lane-change assistance and early warning technology,but there are still some limitations in the current lane-change behavior identification and prediction methods for intelligent and connected vehicles in open scenes.In response to this problem,the dissertation conducts research on the multi-source information fusion method for intelligent and connected vehicle in assisted lane-change technology,focusing on the analysis of lane-change behavior characteristics and parameter extraction,the detection of assisted lane-change traffic information based on deep learning methods,and the design of high-precision assisted lane-change positioning information solution model based on machine learning,as well as the detection and prediction method of multi-source information fusion lane-change behavior.The main research contents and innovations are as follows:(1)The lane-change behavior and feature parameter extraction in the intelligent and connected environment are studied,the feature selection method of vehicle driving state is designed,and a learning-based vehicle lane-change behavior detection model in expressway scene is proposed.Aiming at the intelligent and connected driving environment,combined with relevant information such as driving style and vehicle driving status,the characteristics of vehicle lane-change behavior are deeply researched and analyzed.The parameters related to lane-change behavior are extracted to establish a real scene dataset and annotated.Relevant models are established based on machine learning methods such as random forest and extra tree,and the feature importance of vehicle lane-change behavior related parameters is calculated on the established data set,thereby realizing the selection of the feature parameters of the lane-change behavior prediction model.Based on the selection of feature parameters,a k-nearest neighbor algorithm model is designed to detect the vehicle lane-change behavior,and the detection results of the random forest and extra tree models are compared and analyzed,and the effectiveness of the proposed feature parameters and detection model in vehicle lane-change behavior detection is verified.(2)Based on the analysis of auxiliary lane-change traffic information and traffic object characteristics in real scenarios,vision-based deep learning detection models for traffic object information are proposed.First,for the detection of traffic signs,an improved YOLOv3 deep learning model is designed,Dense Net is introduced to improve the detection effect of small objects in real scenes and the problem of poor clarity caused by light,distance,etc.Then traffic obstacles are classified,relevant data sets are established and labeled,deep learning detection models based on YOLOv3,YOLOv4 and YOLOv4-tiny are designed and established respectively,and the problem of misidentification between obstacles with different degrees of influence of lane-change behavior,improved models based on Wasserstein distance loss function are proposed.Finally,an improved detection model based on YOLOv4 is proposed for the interference problem in vehicle taillight turning information recognition.The proposed model and method are tested and analyzed to verify the effectiveness.(3)On the basis of studying the high-precision positioning and error elimination methods required to obtain the driving state information of auxiliary lane-change vehicles,a GPS/BDS integrated positioning solution method based on LSTM recurrent neural network is proposed.Aiming at the multi-source errors in the positioning process,a positioning solution method is designed and implemented based on the measured data of the GPS/BDS integrated positioning receiver,and an improved positioning solution method is proposed combined with the long short term memory model.The positioning error sequence is predicted and error corrected through a recurrent neural network,and the model is established,thereby improving the accuracy of integrated positioning.Experiments were carried out in static and dynamic scenarios,respectively.The experimental results were compared and analyzed to verify the effectiveness of the proposed method.The lane-change behavior is detected based on the measured data of on-board high-precision positioning,which verifies the effectiveness of the vehicle-mounted high-precision positioning data obtained by the proposed method.(4)On the basis of studying the selection of the feature parameters of the lane-change behavior and the acquisition of the parameters of the driving state around the vehicle,a multi-source information fusion prediction model of the lane-change behavior and its detection method are proposed.Considering the influencing factors such as driving style and the driving state of surrounding vehicles.Based on the selected feature parameters,the vehicle lane-change behavior prediction method based on multi-source information fusion is studied.First,a lane-change behavior detection method based on the vehicle’s driving state information is proposed.On this basis,combined with the driving style and the driving state information of surrounding vehicles,a multi-source information fusion prediction model of lane-change behavior is proposed.A method based on Hidden Markov Model is designed to judge the current lane changing environment,then predict the vehicle driving state parameters,and the lane-change behavior is predicted by the proposed lane-change behavior detection method.Finally,the accuracy and effectiveness of the proposed method are analyzed and verified on the real scene dataset.