Research On LiDAR Modelling And LiDAR-based Driving Threat Assessment

Since world accepted standards for automated vehicles have been set,intelligent vehicle technologies have entered a new stage of development worldwide.With the development strategy “Made in China 2025” in China,Intel igent Connected Vehicle(ICV)signified the significance of intelligent vehicle technologies to our automobile industry.Intel igent vehicle technologies help improving driving safety.During travel ing on the roads,vehicles are expected to give rise to the least chance of collisions or even avoid them.Hence,threat assessment is introduced to estimate and predict potential dangerous situations just like guards for vehicles,which is of great research value.For ICVs,sensors are the bridge between environment information and the vehicles.Sensor information and test scenarios threat assessment confronts is become more and more complicated with the level of vehicle automation rises.LiDAR can conduct distance detection with high resolution,as well as shape detecting,has become a new focus on intel igent vehicle sensors.Thus,it is necessary to research on the application of LiDAR in threat assessment.In the other hand,intelligent vehicle technologies are described as the solution for safe driving problems,so the test and development of technologies also bring dangerous and costs.Simulation technologies is meant to decrease the costs.Building abundant sensor models is indispensable to make simulations validate and integral.This thesis works on the environmental perception based on LiDAR,the threat assessment with the perception result and the simulation of LiDAR in virtual driving environment.The main work of this thesis is as follows:1)LiDAR simulation for virtual driving environmentBoth the detection function and mechanism are considered.LiDARs are modelled in both geometric and physical methods.The mechanism of signal attenuation and noise points caused by rain drops is considered.A simplified LiDAR formula is derived,with easier calibration and high accuracy.Weather influences on signal attenuation are modelled for adaption to dif-ferent weathers.Generation of noise return caused by rain drops is modelled with its mechanism.The models are calibrated and verified with LiDAR data in field tests.2)Vehicle perception based on LiDAR dataLiDAR data are processed for threat assessment,focused on vehicle target perception.Data are collected in dynamic environment with multi-targets on the campus.Region of interests are selected and returns from the ground are segmented.The environment is represented in grid maps.The boundaries of road are extracted,connectedness is applied for vehicle detection.The motions state of vehicles are estimated by Kalman Filter tracker.3)Threat assessment of the traffic environmentThe basic conceptions about threat assessment is described.Maneuver-based trajectory prediction is introduced.The maneuvers of vehicles are recognized with motion states.The kinematic models corresponding to every maneuver is applied for trajectory prediction.Vehicles are represented by bounding boxes,and arbitrary collision detection based on Separating Axis Theorem is applied on predicted trajectories,so that time-to-collision can be calculated efficiently and accurately.A threat level index based on time-to-collision is used to quantize the threat degree of every traffic vehicle potential collision to the ego vehicle.Finally,the accuracy of the calculation of time-to-collision is discussed.4)Simulation and experiment validationThe simulation software PanoSim with the proposed LiDAR model works as simulation platform.The threat assessment method is applied in an autonomous emergency braking system,and tested with scenarios in Euro NCAP test protocols.LiDAR data from filed tests are used for verifying the data processing results and partially validating treat assessment methods.