Research On Comprehensive Cognition Of Traffic Vehicles And Virtual Test For Intelligent Vehicle

After more than one hundred years of development,automotive industry has been developed from the original pure mechanical system to the electro-hydraulic integrated mechanical system,and is moving towards the intelligentialize stage.Many production vehicles equipped with level one or level two driving automation systems have entered the market already,the domestic and foreign researchers are committed to the research,development and productization of level three and above automatic driving vehicles.For the level three and above automatic driving systems,as there is no need for driver to monitor the traffic environment all the time any more,it becomes the necessary prerequisite to cognize the surrounding environment for these high level automatic driving system.Especially in complex traffic scenarios with many vehicles,to cognize the traffic vehicles which is one of the most important factors in complex traffic scenarios,becomes the basic prerequisite for executing all subsequent decision and control method.At the same time,it is inseparable from the scientific,comprehensive,standardized test and certification,before high level automatic driving system really going to the market and being accepted by consumers.However,if the test method still follows the traditional test method for ADAS,at least billions of kilometers of road test is needed to meet the certification requirements.Especially for complex traffic scenarios with many vehicles or extremal conditions,the test faces many difficulties such as the cost,safety and controllability.Therefore,the test and certification method which combinate the virtual test--field test--road test is becoming the hotspot to research.The virtual test based on driving simulator has became an indispensable part for intelligent vehicle virtual test,because it have not only the advantages of economic,safe and controllable for test conditions,but also can conduct subjective and objective evaluation with real driver in the loop.Duiring cognizing the traffic vehicles: the relative motion between vehicles will make occlusion to each other,which results in that sensors cannot percept the occluded vehicles;to reduce the sensing system’ cost for intelligent vehicles,multi-sensor system layout on the vehicles may form some unavoidable blind spots,so that the system cannot obtain the perceptual result for traffic vehicles in the blind spots;the existing sensors normally can only percept partial traffic vehicles’ motion,and with false or missing sometimes.Therefore,the cognition of traffic vehicles is not only the problem of improving accuracy by multi-sensor data fusion at one single moment,it should track the traffic vehicles in the whole sensing scope steadily and estimate their motion state accurately all the time.During testing intelligent vehicles in virtual: even for a single intelligent vehicle simulation,the physical sensor model based on physical and mechanism has still not completely conquered the problem of real-time simulatin,because its complex calculation and high requirements for scene modeling;the functional sensor model based on statistical characteristics is relatively simple to model because it ignores the internal working process,but that’s not enough for testing performance of comprehensive cognition method and other high level automatic driving system,it still needs additional complex calculation to simulate other necessary physical phenomenon,so it is also very difficutl to support multi-intelligent vehicles’ real-time simulation in a complex scenario.Therefore,to bulid an appropriate sensor model with necessary physical phenomenon and high computational efficiency,is still a technical problem for intelligent vehicle’s virtual test to solve.Contraposing the above problems on cogizing traffic vehicles and virtual test for intelligent vehicle,this paper tries to explore a traffic vehicles’ cognition mothed based on maneuvering target tracking theory,which can track and estimate traffic vehicles even when sensing information is missing or obvious mistaken for a short time;to test the cognition method using dirving simulator,a functional sensor model with necessary physical phonomemnon and high computational efficiency is built,it’s also suitable for simulating complex testing scenario where there are many intelligent vehicles;this paper also studies the key techniques for integrating intelligent vehicle simulation-oriented virtual test platform on driving simulator.The main research contents as follows:Firstly,for the cognizing of traffic vehicles: Contraposing the problems of data losing and mistake percept during cogizing traffic vehicles,this paper propose a comprehensive cognition method for traffic vehicles based on maneuvering target tracking theory,and explore the key method for tracking and estimating.Including: a data association method which fuses diffirent objects’ feature based on DS evidence theory and collision detectiong algorithm,this method is effective even if sensors percept one big vehicle as several objects;a meothod to estimate traffic vehicle’s geometry profile by fusing the profile feature and classification is proposed;a method to estimate traffic vehicle’s motion state based on adaptive filter algorithm is studied.The road experiment results show that the proposed comprehensive cognition method can track different kinds of vehicles steadily on the urban expressway and the urban ring road,and some qualitative analysis is shown about the accuracy of the estimated vehicle’s geometry profile and relative and absolute motion state.Secondly,for simulating the function of sensors: Most of existing functional sensor models can only reflect perceptual error without other common phenomenon which result from the physical character,so it’s not suitable for tesing comprehensive cognition method and other high level automatic driving systems,if consider other physical phenomenons,the computational load becomes high too,so it’s either not suitable for simulating the complex traffic scenario with many intelligent vehicles in real time.Aiming at this,a functional sensor modeling method is proposed which can reflect both error and other nessecary physical phenomenon with high computational efficiency,and the key technical methods are studied.In order to realize simulating complex traffic scenario with many intelligent vehicles,this paper proposes a quick culling method based on spatial location and object’s geometry profile,which can effectively manage and efficiently retrieval the large scale 3D simulation scene data.To simulate oher common physical phonomenons which result from the sensors’ physical character,a fast occlusion relationship deciding algorithm and a fast algorithm for lidar percepting partial geometry profile are proposed,the two algorithms are both based on the visible triangle and visible angle which are defined in this paper.The simulation results show that: the radar model,lidar model,camera model based on the method in this paspr can reflect the product characteristics well and computational efficiency is also high enough to simulate many intelligent vehicles in a complex scenario in real time.Thirdly,in order to integrate the intelligent vehicle simulation-oriented virtual testing platform based on driving simulator,this paper analysis the new demand for intelligent vehicle simulation to driving simulator comprehensively,and some key technique for integrating the platform are studied.Including: a layered method to model road based on road segment,it models the road structure and logical connetion information and so on for traffic and sensor simulating on the traditional 3D road model which focus mainly on the geometry and illumination reality;an integrated framework for building up distributed vision system to ensure the real-time display effert with a large scale of data to update when traffic simulator is connected to driving simulator;and a correction method to follow the motion system for vision system for opening screen driving simulator,which ensures the accordance between vision and motion systems,and is an important improvement for driving simulator to evaluate automated driving system’s performance subjectively and objectively.This series of methods have been applied in different scale and structured driving simulators,and these driving simulators are being applied for subjective and objective evaluation for intelligent vehicle’s control performance.Finally,a hardware platform for the comprehensive cognition mothed,and a software development platform which is generalduty on both real vehicle and driving simulator are integrated.The comprehensive cognition method proposed in this paper is tested on both driving simulator and real vehicle,the test results show that: the comprehensive cognition method can track traffic vehicles steadily and estimate their motion state accurately in the whole cognition scope for intelligent vehicle;the functional sensor modeling method and the key technique to integrate driving simulator for testing intelligent vehicles can be applied to the intelligent vehicle product development and virtual test.