Research On Vehicle Driving Behavior Recognition And Trajectory Prediction Method

Intelligent vehicles have an intrinsic essense of safety,efficiency,comfort and energy saving on account of equipment with advanced automation and information technology.Therefore,the intelligentization of vehicles has gradually become the future development trend of the vehicle industry.However,with the development of intelligent driving technology,intelligent vehicles are showing a process of gradual development as well as immense popularization,where complicated scenes like fully automatic driving vehicles,semi-autonomous driving vehicles and manual driving vehicles coexisting on the road can be expected.Due to complexity and stochasticity of driving behavior and driving trajectory of the target vehicle,the perception of the current traffic environment alone is not enough to ensure the safe driving of the intelligent vehicle.Comprehension of driving behavior of the dynamic target vehicle and prediction of its future driving trajectory are necessarily required.In this paper,research on the recognition of target vehicle driving behavior and prediction of driving trajectory in a dynamic environment is conducted in order to improve the ability of intelligent vehicle environment perception and understanding,and the performance of the algorithm is verified and analyzed through data collected in real vehicle experiments and a simulation platform.Aiming at the Mobileye Eye Q3 camera sensor’s limit ability to provide the lane line information in front of the vehicle,where there are crossovers outside the effective distance of the forward lane line,this paper proposes a lane line forward correction algorithm and a rear lane line estimation algorithm.First,based on the structured road characteristics,the forward lane line is described in segments,which corrects the problem of lane line crossing outside the effective distance,and improves the accuracy of long-distance lane line.Then,it is proposed to use multi-frame historical time lane lines as input,and use the principle of vehicle coordinate transformation combined with the idea of least squares to complete the calculation of the rear lane line,which makes up the lack of the rear lane line in the environment perception and provides strong evidence for rear lane line positioning as well as driving behavior recognition.Due to flexibility and uncertainity in the driving behavior of the dynamic target vehicle in the road environment,the correlation between the driving behavior of the vehicle and the Markov process is analyzed.The driving behavior is divided into three stages: preparation,execution,and termination.The GMM-HMM driving behavior recognition model was established for left lane change,right lane change and lane keeping,and real vehicle experiments were used to collect data,and the driving behavior recognition model was trained and verified.The statistical analysis results show that the method can achieve timely and accurate recognition of driving behavior,enhance the environment perception ability,and provide a guarantee for trajectory prediction and intelligent decision-making planning.Considering that the single trajectory prediction method has its own problems and cannot achieve the perfect representation of the predicted trajectory,based on model interaction,a long and short time-domain trajectory prediction method is proposed,which combines driving behavior and motion status to realize the detection of vehicles in the future prediction of the trajectory.This method not only guarantees the prediction accuracy in the short-term domain,but also ensures that the prediction is consistent with its driving trend in the long-term domain.After simulation and verification,the lateral error of the fusion algorithm is kept within 0.2m in the prediction time of up to 5 seconds,which shows the effectiveness of the trajectory prediction algorithm,improves the understanding of the intelligent vehicle to environmental changes,and provides convincing evidence for intelligent decision-making.This paper aims to improve the understanding and prediction of dynamic environment changes of intelligent vehicles.Around the intelligent vehicle environment perception technology,this paper proposes lane line related algorithms,target vehicle driving behavior recognition model and its driving trajectory prediction algorithm.The experimental results show that:(1)the problem of missing lane lines in the rear and wrong lane lines in the front is effectively solved;(2)the driving behavior of the target vehicle can be effectively identified in a timely and accurate manner,high trajectory prediction accuracy can be acquired.The research in this paper not only improves the environment perception ability of intelligent vehicles,but also provides a solid basis for intelligent decision-making and control,and has high practical application value.