文章编号:1009−444X(2022)04−0392−06基于激光和视觉传感器融合的定位与建图赵以恒,周志峰(上海工程技术大学机械与汽车工程学院,上海201620)摘要:定位与建图是自动驾驶的关键技术之一.激光传感器或视觉传感器具有局限性,通过多传感器融合可以发挥不同传感器各自的优点,提高定位与建图的精度和鲁棒性.通过优化Harris算法对图像进行角点提取,利用关键帧对特征点匹配算法进行优化,然后利用非线性最小二乘法进行后端优化.通过试验平台进行定位与建图试验,对算法进行验证,并用EVO工具对定位误差进行分析.结果表明,提出后端优化算法误差比单一传感器定位误差减少13%.关键词:多传感器融合;角点检测;特征点匹配;非线性最小二乘法中图分类号:TP242文献标志码:ALocationandmappingoflidarandvisionsensorfusionZHAOYiheng,ZHOUZhifeng(SchoolofMechanicalandAutomotiveEngineering,ShanghaiUniversityofEngineeringScience,Shanghai201620,China)Abstract:Locationandmappingisoneofthekeytechnologiesforautonomousdriving.Withlimitationsoflidarsensorsorvisionsensors,theadvantagesofdifferentsensorscanbebroughtintoplaythroughmulti-sensorfusionandtheaccuracyandrobustnessoflocationandmappingcanbeimproved.TheHarrisalgorithmwasoptimizedforcornerextraction,thekeyframewasusedtooptimizethefeaturepointmatchingalgorithm,andthenthenonlinearleastsquaremethodwasusedforback-endoptimization.Thelocationandmappingexperimentswerecarriedoutonthetestplatformtoverifythealgorithm,andthepositioningerrorwasanalyzedwiththeEVOtool.Theresultshowsthattheerroroftheproposedback-endoptimizationalgorithmis13%lessthanthatofasinglesensor.Keywords:multi-sensorfusion;cornerdetection;featurepointmatching;nonlinearleastsquaremethod移动机器人近几年得到较快的发展,精准的定位和准确的地图对于移动机器人非常重要.利用自身携带的传感器,通过对周围世界的观测和自身运动的解算,来完成自主定位与建图,这便是同步定位与建图技术(SimultaneousLocationandMapping,SLAM).在SLAM系统中,常见的传感器包括激光传感器和视觉传感器.多线激光雷达稳定性较高,但点云特征不明显,不易对障碍物进行很好的分类.相机传感器的成本低,便携程度好,提供丰富的环境信息,但在夜晚或者光线较弱的情况下,识别能...