Research On Lane-changing Intention Recognition Method For Freeway Driver

Owing to the high speed, the loss of life and property caused by traffic accidents inhighway is relatively more serious. Lane changing and keeping are two typical drivingbehaviors of highway, which have serious impact on driving safety. Therefore, variouslane-changing assistant systems have been developed by researchers. When perceivingenvironment, the existing systems mainly use radar and vision for environment informationand vehicle state perception and always neglect driver’s behavior motivation and drivingintention. So due to the misunderstanding of driver’s true intention, systems often alarm andexecute actions forcibly dissimilarly with driver intention, which could lead to driver’sdistraction, tension even losing regular control of vehicle, then the active safety systembecomes the incentive for traffic accidents. As a result, intention recognition of driver’slane-changing in highway environment has important significance for enhancing drivingsafety and improving traffic environment.In specific traffic environment, driver’s behavior has a certain regularity and similarity.As an ego inner state, driver intention couldn’t be obtained directly during driving process,but conjectured through indirect information such as driver’s action, gesture and vehiclestate during the driving. Simply, drive intention determines the behavior and drive behaviorrealizes the intention. Analyzing and deducing the driver’s visual characteristics and drivingperformance measures is an important method for driver intention recognition.This research was supported by the National Natural Science Foundation for YoungScholars of China “Research on driving behavior safety warning system for lane-changingand overtaking”. By summarizing current research achievement at home and aboard, withthe aim of recognizing lane change intention on highway situation accurately and timely, thisresearch went into the driver’s visual characteristics and vehicle performance measureschanging law during the stage of lane-changing intention and lane-keeping from drivers withdifferent driving styles. As a result, the effective characteristic parameters set of lane changeintention was got optimized. Lane-changing and keeping intention recognition model forhighway were researched and established. Finally, the effect of driving styles, characteristicparameters and modeling methods on intention recognition was analyzed and compared according to accuracy, sensitivity and specificity. The achievements lay the theoreticalfoundation and technical support for the pratical lane change intention recognition system.The specific research contents are as follows:1. Experiment program designation and data collection. First, the lane-changingbehavior was divided into intentional and executive section based on the related research athome and abroad combined with this research’s purpose. Second, experiment project wasdesigned, including installing the collecting devices, developing the traffic scenes,standarding the experiment process, selecting the measuers, etc. Based on questionnaires,21drivers were divided into radical, conservative and normal according to their driving styles.Through removing the abnormal data and combining the lane-changing intention timewindow for different style drivers, trainning and test database for different styles drivers andintentions was established, which could provide data support for following research.2. Law analysis of drivers’ visual characteristics in different intention stages. First, thegaze areas in driving was divided into forward、left rear-mirror、right rear-mirror、instrumentpanel and internal rear-mirror by means of combing static and dynamic state based on drivercharacteristics. Second, from the perspective of gaze behavior、glance behavior and headrotation, combining with experimental data and videotape, the change law of visualcharacteristic parameters of different style drivers in lane-changing intention andlane-keeping stages were analyzed deeply, and independent-samples T test (T-test) andanalysis of variance (ANOVA) were used to quantify the effects of driving stage and drivingstyle on visual parameters. Finally, the optimal visual characteristics parameters set whichcould characterize the lane-changing intention was the fixation number of rear-mirror，theaverage saccade amplitude and the standard deviation of head horizontal direction angle.3. Law analysis of driving performance measures in different stages. First, turn signalopen rate and open time in advance of different style drivers were statistically analyzed.Second, with classifying driving performance measures by lateral and longitudinalmovement, driving performance measures of different style drivers in lane-changingintention and lane-keeping stages were analyzed deeply. Based on the visual analysis ofessential features, the independent-samples T test (T-test) and analysis of variance (ANOVA)were used to quantify the effects of driving stage and driving style on driving performancemeasures. Besides, the research studied the correlation between driving performancemeasures by taking advantage of correlation analysis, and provides the basis for reducing thedimensionality of the multidimensional feature space. Finally, steering entropy was choosenas the driving performance characteristics parameters which could reflect the lane-changingintention in highway. 4. Research on the driver intention recognition model. According to the visual anddriving performance characteristics parameters of driver lane-changing intention in highway,driver intention recognition models were developed separately based on GM-HMM theoryand SVM theory. Different types of characteristics parameters were used respectively to trainthe driver intention recognition models off-line for constrasting the effect under differentcharacteristics parameters, and then the relevant model parameters were obtained.5. Model evaluation and recognition effect analysis. Tested the established models usingthe test set, and evaluated the models by utilizing three model evaluation functions whichincluded accuracy, sensitivity and specificity，and the effect of driving styles, characteristicparameters and modeling methods on intention recognition was analyzed deeply. The resultsshow that the optimal recognition effect is achieved when choose Vehicle&Head&Eye asthe observation sequences and develop the recognition model based on GM-HMM theory.Lane-Changing intention recognition methods and theories on highway were studieddeeply in this paper. The research analyzed that the impact of driving styles on drivingbehavior and established characteristic parameters which reflect lane-changing intentioneffectively. Finally, reasonable intention recognition model was set up. The research resultsprovide theoretical and technical support to the related field of active safety assistancesystems, and improve traffic safety, comfort and the traffic environment.