Research On Driving Behavior Prediction Method Based On Driver’s Visual Characteristics

The traffic environment of urban road is complicated, and traffic conflicts occurfrequently between vehicles, wrong driving behavior will lead these conflicts to bemore serious and cause traffic accidents. Because the trend that drivers areunprofessional and traffic is intensive become apparent increasingly in urban trafficflow recently, the research on driving behavior in urban traffic environment isparticularly important. And because driving intention is the internal state of drivingbehavior, the study on driving intention is the premise of driving behavior prediction.So driving intention identification and behavior prediction in urban trafficenvironment is significant to improve driving safety and traffic environment.This paper analyze and summarize the driving intention recognition and behaviorprediction methods that are existing at home and aboard, driving intention recognitionis selected as the bridge and predict driving behavior of the urban traffic environment.According to research and analysis driver’s visual characteristic parameter in differentdriving behavior and intention of car following, overtaking, braking andlane-changing (including left lane-changing and right lane-changing) thoroughly,select typical visual characterization parameters to characterize driving behavior andintentions, and propose a driving behavior prediction method based on driver’s visualcharacteristics by establishing driving behavior prediction model. These provide thetheoretical basis and technical support for the practical of driving behavior warningsystem.The study contents are summarized as follows:1. Urban Traffic driving behavior and driving intent analysis. Firstly, byanalyzing the driver’s behavior under urban traffic environment, to divide the driver’sintentions under the urban environment into4class, car following, overtake and theleft and right lane changing; Secondly, by analyzing the relation between the driver’sbehavior and intention, to confirm the prediction method of driving behavior based onthe driving intention recognition. With the target of the driving behavior prediction,analyze the impact factors of driving intention and behavior, confirm the visualfeature parameters of driving intention and behavior, and all above would lay the foundation for the driving test design and the selection of driving behavior predictionparameters.2. Test designation and data collection. Firstly, the real vehicle road test isdesigned, including the analysis of the test purpose, the selection, installation anddebugging of the test equipment, the selection of the personnel and route, the testprocess development and so on. Secondly, through the road test, collect the real-timedata of the driver behavior and visual characteristics; thirdly, eliminate abnormal databy using the PauTa, and analyze the driver behavior through the test video, and thenselect the driving behavior sample. Finally, select the reasonable data for the drivingbehavior prediction, this paper use4s,4.5s and5s data as the training sample andevaluating sample database for driving behavior prediction, and it also provides datasupport for the construction of driver behavior prediction model.3. Research on the method of driver’s vision plane divided based on the driver’sfixation points distribution. Firstly, use k means clustering and select k6， k7and k8, cluster driver’s fixation points. Secondly, to driver’s abnormal fixationpoints are removed used the guidelines of PauTa. Then, the driver’s fixation pointsafter clustering will be projected onto the front wind glass of the vehicle. Throughcomparative analysis of the advantages and disadvantages of each sub-cluster,determine the final clustering results. Finally, based on the distribution of the driver’sfixation points, divide the driver’s vision plane into seven parts: front lane, left lane,right lane, left rearview mirror, right rearview mirror, interior mirrors and cardashboards.4. Law analysis of driver’s visual characterization parameters in different drivingbehavior. Firstly, from the three aspects of fixation, saccade and rotation of head,combining with the test video and test data, to deep analyze the driver’s differentdriving behavior that correspond to visual features and the change rule of headmovement characteristics parameters, and use multivariate analysis of variance tostudy the influence on driver visual parameters from different driving intentions. Then,determine visual indicator that could represent driver’s driving behavior.5. Establishment of driving behavior prediction model. Firstly, Use the basic idea,arithmetic and classification of the theory of hidden Markov to elaborate the HMMmodeling idea; Then according to the research purpose and requirement, select thevisual characterization parameters needed for modeling driving behavior, and select100samples as the training sample from all driving behavior to train the model, andconstruct the driving behavior prediction model on the basis of HMM arithmetic; Through the correlation analysis, the selected characteristics can be simplified, andthe fixation count, saccade duration, saccade speed, Fixation point transitionprobability and rotation of head are selected as the modeling parameters.6. Model evaluation and prediction result analysis. Firstly, analyze the predictionresult through calculating the accuracy of predict model. It shows that the accuracy ofprediction models can reach more than85%when using4.5s as the time window toselect prediction sample; Then the paper studied the impact when using4s or5s as thetime window, it shows that it is appropriate to use4.5s as the time window to predictdriving behavior on the purpose of dangerous driving behavior early-warning.This paper systematically and deep studies the key technology problems aboutdriving behavior prediction and intention recognition system based on visualcharacteristic on urban road. By analyzes the influence factors of urban trafficenvironment on drivers and drivers’ driving intentions, establishes visualcharacteristic parameters that represent driving behavior on urban road, andestablishes driving behavior prediction model. These provide the theoretical basis andtechnical support for the practical of driving behavior warning system and applicationof vehicle active safety assistant system, thus to improve the driving safety andcomfort of drivers and improve the traffic environment.