Research On Man-machine Co-driving System Based On Intention Recognition

In recent years,the improvement of on-board ECU computing power has provided conditions for the birth and application of Advanced Driving Assistance System(ADAS),gradually becoming a standard feature of new car models.ADAS can effectively reduce the probability of misoperation caused by drivers’ misjudgment,distraction,and other reasons,improve driving safety,and reduce the burden on drivers.However,ADAS decisionmaking cannot guarantee 100% accuracy,and correct decision-making may also reduce the driving experience of drivers and passengers due to differences in their psychological expectations.Due to various factors such as current technological development and legal division of responsibilities,fully autonomous driving still needs time.Therefore,accurately identifying the driver’s driving intention and using it as the basis for ADAS decisionmaking to improve the accuracy of ADAS decision-making is an important direction for the future development of ADAS.This article first conducts data collection experiments.Using a driving simulator as a platform,set up experimental vehicles and scenarios in Car Sim software,and provide vehicle motion state parameters;Install optical equipment and use deep label distribution learning to detect head posture.To approach the real driving experience,hardware such as the steering wheel,accelerator,and brake pedal are installed,which are detected by sensors and transmitted to the host computer in real-time through the target machine,achieving hardware in the loop simulation.Recruit experimental personnel,standardize experimental procedures,and collect experimental data.Subsequently,an online recognition system for driving intent was established.Screen lane change feature parameters,which are divided into vehicle lane change intention parameters and driver observation intention parameters based on the characteristics of vehicle movement and driver operation;The lane changing behavior is determined by the lateral position of the vehicle’s centroid,and the lane changing sections are divided based on the different mutation time points of the two types of parameters before the lane changing.A driving intention recognition model based on the Conditional Random Field(CRF)is established.The training is carried out with the collected experimental data as samples.The three driving intentions of left lane change,right lane change and lane keeping are recognized offline.Compared with the Hidden Markov Model,CRF has advantages in accuracy and recognition speed,and the time of single recognition intention is less than the data sampling time.Randomly select a long sample to test the recognition effect of multiple driving conditions,proving the stability of the model in identifying driving intentions.On the basis of offline recognition,a driving intention online recognition system is established,and the driving intention recognition code is embedded in Simulink,so that the driving intention recognition results and vehicle motion status can be displayed in real-time on Veriland’s workspace,achieving CRF based driver intention online recognition.By comparing the online recognition results with the offline recognition results,it is proven that the system can stably output the driver’s driving intention.Finally,a man-machine co-driving decision-making system is established.The manmachine co-driving mode was determined,and the Starkelberg game model was established based on the driver’s intention parameters and vehicle collision warning data.The maximum safety distance was obtained by game with the following vehicle,and the benefit of lane change decision was comprehensively analyzed by combining the changes of driver’s intention and acceleration space.Simulation experiment scenes are set up to verify the effectiveness of the system.The research results of this article provide reference for the development of driving intention recognition and human-machine co driving decision-making.