Drivers' Lane Changing Intention Recognition Method Research Based On Hidden Markov Model

In recent years,with the continuous improvement of the intelligence of vehicle,the intelligent driving assistant system comes into being.Most of the current intelligent driving assistant systems are based on the intelligent perception of the vehicle sensors,and then carry out intelligent planning,decision-making and execution.In this process,if the intention of the main driver of the driving task is ignored,it may lead to the decision of the driver contrary to the intelligent driving assistant system,so there will be more potential accident risks.Respecting the driver's intention can not only better meet the human's pursuit of driving safety and comfort,but also improve the acceptance of intelligent assistant driving system,which is also a necessary part of improving human-computer co-driving.In order to accurately identify the driver's driving intention,this paper focuses on the analysis of the driver's psychological and physiological behavior in the process of lane changing,and proposes a method of lane changing driving intention recognition based on the Gaussian Mixture Hidden Markov model.Based on the analysis,summary and study of the existing research results of driver intention recognition and behavior prediction at home and abroad,this paper combines major project of NSFC "active vehicle safety collaborative control and application verification under extreme conditions"(No.61790560),national key R &D plan "personalized human machine co driving theory based on driver driving skills"(no.2016YFB010904),key project of NSFC regional innovation and development joint fund "intelligent driving decision and human vehicle cooperation in ice and snow environment" Research on Key Technologies of the same control(No.U19A2069),special project "advanced technology of chassis electrification of next generation passenger vehicles"(No.SXGJSF2017-2-1-1)of Jilin Province University co construction plan and youth fund project "Research on active obstacle avoidance control method of CO driving intelligent vehicles respecting driver's intention"(No.661790560),so as to accurately identify the driver intentionrecognition and behavior prediction aiming at the driver's intention of changing lanes,this paper analyzes the driver's visual characteristics and vehicle state in the generation stage of the intention of changing lanes,selects the data combination of the characteristic variables that can represent the driving intention,establishes the Hidden Markov model of driver's intention recognition of changing lanes,and analyzes the recognition of the intention of changing lanes by various characteristic variables from the perspective of probability accuracy.This paper mainly contains the following aspects:1.Real vehicle experiment design and data acquisition.In order to accurately obtain the data that can reflect the driver's driving intention,this paper designs the driving experiment of changing lanes,collects the facial video recording,vehicle data and environment information of different drivers in the process of changing lanes,and takes the data information as the basis of model building.2.Feature variable extraction of driving intention.In order to obtain the feature variables for the establishment of intention recognition model,face recognition and eye location are carried out based on the skin color segmentation technology in YCbCr space and the method of ASEF filter,so as to achieve the purpose of quantifying the head and eye movement information of drivers.At the same time,the steering entropy is used to represent the driver's psychological factors when changing lanes.In addition,the vehicle state data and environmental information are referred to as the characteristic variable group of intention.3.Based on the Gaussian Mixture Hidden Markov model,the intention recognition model of lane changing is established.In view of the fact that driving intention is random and unmeasurable,a Gaussian Mixture Hidden Markov model is used to model driving intention recognition.In order to select the main variables of the selected feature variables and train the model,the data of multiple drivers in the process of lane changing are used to train the model based on different variables and test them respectively,and the accuracy of the test is taken as the standard to judge the selection of feature variables.The results show that the combination of feature variables(face/eye information,steering entropy,vehicle and environment state)can be used as the feature variable group to represent driving intention,and the Hidden Markov model trained by the combination of feature variables can be used as the model to predict the driver's intention to change lanes.4.Improvement of driving intention recognition model and open-loop test.In view of the situation that the driver has the intention of changing lanes but there is nointention of changing lanes,the training set of the model is modified to improve the driving intention recognition model by using the data in this case;considering that different drivers have different driving characteristics,the driving intention recognition model is personalized by combining the support vector machine and the Gaussian Mixture Hidden Markov model.Finally,in order to test the application ability of the model,the open-loop test method is used to further evaluate the recognition accuracy and practicability of the model.