基于多帧点云动态配准的激光雷达目标检测方法研究.pdfVIP

Abstract

Inrecentyears,withtherapiddevelopmentofautonomousdrivingtechnology,thetopic

ofenvironmentalperceptiontechnologyonautonomousdrivingreceivedalotofattentionfrom

researchers.Inordertomakeautonomousvehiclesmoreintelligentandsafeduringdriving,

multiplesensorsareoftenintegratedonthevehicle.However,duetothewidedetectionrange,

highrangingaccuracy,andunaffectedbyexternalfactorssuchaslight,LiDARiswidelyused

inautonomousdrivingandisanindispensableandimportantcomponentofenvironmental

sensingsystemobjectdetection.

ThekeypointofLiDarobjectdetectionfocusonhowtodetectobjectaccuratelywithin

lesstime.Onthesametime,thefarthertheobjectis,thebetterLiDarobjectdetectionwillbe.

Thiswillgivestheautonomousdrivingmoretimetomakedecisionandcontrolthevehicle,

increasingtheauto-drivingsafety.Inthispaper,weproposedanobjectdetectionsystem

basedonthefusionofLiDARandInertialMeasurementUnit(IMU)fordynamicregistration

ofmultiframepointclouds.Themaincontributionaslistasfollws:

(1)Toresolvetheproblemoflowreal-timeperformance,ncompletegroundpointcloud

segmentation,andevenoversegmentationphenomenaontraditionalgroundpointcloud

segmentationalgorithm,thethesispurposedastatisticalfilteringadaptiveslopethreshold-based

algorithm.Thisalgorithmdividedthedenoisedpointclouddataintomultiplesectorregionsset.

Theglobalthresholdasdividedintosamesectionnumberofthepreviousdataset,then

adaptivelyselecttheminimumvalueasthelocalthresholdforgroundpointcloudsegmentation.

AfterusingthepurposedalgorithmsonSemanticKITTIdataset,theresultshowthatitcan

achievehigheraccuracyandrecallincomplexscenarios.

(2)Toadd