A Study Of Lane-changing Warning Strategy Based On Semantic Analysis Of Driving Behavior

Describing behavior bus driver behavior accurately is critical to driving safety strategies.Based on collected naturalistic data of six bus drivers,the research contained the following parts:(1)Obtained and preprocessed multi-scenario lane-change data.First,on-board equipment was used to obtain environment data and operating data to build a database.Then,driving behavior parameters were decoded and analyzed.In addition,typical approaching scenes and lane-changing scenes of bus drivers were defined,along with rules being designed to extract specific scenarios.After that,data preprocessing was carried out,this has laid a foundation for bus driving behavior characteristic analysis,driving style clustering and lane change warning strategy verification;(2)Statistical analysis was applied to analyze the tendency of bus drivers in lane change scenes.First off,car following scene and lane change scene were established to analyze the maneuvering characteristics of the bus driver with variance analysis and scatter diagram.Besides,parameters like time headway(THW)and time to collision(TTC)were calculated to see if the distance between the leading and following vehicle were long enough to satisfy the safety standards.Moreover,research about the driver’s decision in car-approaching scene was developed.(3)Several typical lane change scenes were defined and characteristics of bus drivers were concluded.Three criteria which refers to the existence of a vehicle ahead,the initial speed of a vehicle,as well as lane change direction were defined to classify six typical lane change scenarios.On top of that,variance analysis,cumulative frequency calculation and scatter diagram analysis were introduced to facilitate the following research.Overall,speed change analysis in the lane change process,driver’s manipulation difference,state parameters analysis for ego vehicle and the vehicles ahead all contributed to help conclude maneuvering feature of bus driver in the lane change scenes.The conclusion subsequently better guide the construction of personalized driver model;(4)Driving styles(DS)identification method were developed.Based on the driver’s operating parameters and vehicle state parameters,features that are highly relevant to DS were extract as original features.To begin with,principal component analysis(PCA)was used to reconstruct the original features,then iterative self-organizing data analysis(ISODATA)algorithm was adopted to cluster distinct driving styles.Then,K-means,Gaussian mixture model and fuzzy C-means clustering algorithm were selected for comparative analysis to verify the effectiveness and superiority of ISODATA clustering algorithm.(5)A lane change warning strategy based on the minimum safe distance was developed.Several modules like personalized driver module and style identification module were assembled to build a personalized lane change warning system simulation platform.The lane change warning strategy was designed based on the relative distance and relative speed of the leading and the following vehicle.Additionally,a simulation verification platform,which can test the function of lane change data collection,driving style identification online,driver model update and lane change warning by using naturalistic data to verify the fidelity of the proposed method.