Al-5Cu-4.5Mg-2.5Zn alloy was prepared and the alloy ingots were fabricated by squeeze casting in this work.Considering these negative effects of composition segregation and coarse second phases,some heat treatments we...Al-5Cu-4.5Mg-2.5Zn alloy was prepared and the alloy ingots were fabricated by squeeze casting in this work.Considering these negative effects of composition segregation and coarse second phases,some heat treatments were adopted in this research.Microstructures,element distribution,phase constitutions and mechanical properties of Al-5Cu-4.5Mg-2.5Zn alloy ingots before and after heat treatments were investigated.It was discovered that these heat treatments would influence and extremely optimize the microstructures and properties of Al-5Cu-4.5Mg-2.5Zn alloy.Except some residual S (Al_(2)CuMg) phase and a few of η phase,the precipitate free zone (PFZ) and the Guinier Preston zone (GPZ) formed in the alloy.It was also found that θ′′ (Al_(2)Cu) and η′′ (MgZn_(2)) phases formed and kept a consistent relationship with the aluminum matrix.As the result,these properties of ultimate tensile strength (UTS),percentage of elongation and Brinell hardness (HB) were greatly elevated.The UTS,percent of elongation and HB were 469 MPa,8.1% and 208 N/mm^(2),respectively.展开更多
目的激光雷达在自动驾驶中具有重要意义,但其价格昂贵,且产生的激光线束数量仍然较少,造成采集的点云密度较稀疏。为了更好地感知周围环境,本文提出一种激光雷达数据增强算法,由双目图像生成伪点云并对伪点云进行坐标修正,进而实现激光...目的激光雷达在自动驾驶中具有重要意义,但其价格昂贵,且产生的激光线束数量仍然较少,造成采集的点云密度较稀疏。为了更好地感知周围环境,本文提出一种激光雷达数据增强算法,由双目图像生成伪点云并对伪点云进行坐标修正,进而实现激光雷达点云的稠密化处理,提高3D目标检测精度。此算法不针对特定的3D目标检测网络结构,是一种通用的点云稠密化方法。方法首先利用双目RGB图像生成深度图像,根据先验的相机参数和深度信息计算出每个像素点在雷达坐标系下的粗略3维坐标,即伪点云。为了更好地分割地面,本文提出了循环RANSAC(random sample consensus)算法,引入了一个分离平面型非地面点云的暂存器,改进复杂场景下的地面分割效果。然后将原始点云进行地面分割后插入KDTree(k-dimensional tree),以伪点云中的每个点为中心在KDTree中搜索若干近邻点,基于这些近邻点进行曲面重建。根据曲面重建结果,设计一种计算几何方法导出伪点云修正后的精确坐标。最后,将修正后的伪点云与原始激光雷达点云融合得到稠密化点云。结果实验结果表明,稠密化的点云在视觉上具有较好的质量,物体具有更加完整的形状和轮廓,并且在KITTI(Karlsruhe Institute of Technology and Toyota Technological Institute)数据集上提升了3D目标检测精度。在使用该数据增强方法后,KITTI数据集下AVOD(aggregate view object detection)检测方法的AP_(3D)-Easy(average precision of 3D object detection on easy setting)提升了8.25%,AVOD-FPN(aggregate view object detection with feature pyramid network)检测方法的AP_(BEV)-Hard(average precision of bird’s eye view on hard setting)提升了7.14%。结论本文提出的激光雷达数据增强算法,实现了点云的稠密化处理,并使3D目标检测结果更加精确。展开更多
基金Funded by the International Cooperation Project of the Ministry of Science and Technology of China (No. 2014DFR50320)the National Natural Science Foundation of China (No. 51174064)。
文摘Al-5Cu-4.5Mg-2.5Zn alloy was prepared and the alloy ingots were fabricated by squeeze casting in this work.Considering these negative effects of composition segregation and coarse second phases,some heat treatments were adopted in this research.Microstructures,element distribution,phase constitutions and mechanical properties of Al-5Cu-4.5Mg-2.5Zn alloy ingots before and after heat treatments were investigated.It was discovered that these heat treatments would influence and extremely optimize the microstructures and properties of Al-5Cu-4.5Mg-2.5Zn alloy.Except some residual S (Al_(2)CuMg) phase and a few of η phase,the precipitate free zone (PFZ) and the Guinier Preston zone (GPZ) formed in the alloy.It was also found that θ′′ (Al_(2)Cu) and η′′ (MgZn_(2)) phases formed and kept a consistent relationship with the aluminum matrix.As the result,these properties of ultimate tensile strength (UTS),percentage of elongation and Brinell hardness (HB) were greatly elevated.The UTS,percent of elongation and HB were 469 MPa,8.1% and 208 N/mm^(2),respectively.
文摘目的激光雷达在自动驾驶中具有重要意义,但其价格昂贵,且产生的激光线束数量仍然较少,造成采集的点云密度较稀疏。为了更好地感知周围环境,本文提出一种激光雷达数据增强算法,由双目图像生成伪点云并对伪点云进行坐标修正,进而实现激光雷达点云的稠密化处理,提高3D目标检测精度。此算法不针对特定的3D目标检测网络结构,是一种通用的点云稠密化方法。方法首先利用双目RGB图像生成深度图像,根据先验的相机参数和深度信息计算出每个像素点在雷达坐标系下的粗略3维坐标,即伪点云。为了更好地分割地面,本文提出了循环RANSAC(random sample consensus)算法,引入了一个分离平面型非地面点云的暂存器,改进复杂场景下的地面分割效果。然后将原始点云进行地面分割后插入KDTree(k-dimensional tree),以伪点云中的每个点为中心在KDTree中搜索若干近邻点,基于这些近邻点进行曲面重建。根据曲面重建结果,设计一种计算几何方法导出伪点云修正后的精确坐标。最后,将修正后的伪点云与原始激光雷达点云融合得到稠密化点云。结果实验结果表明,稠密化的点云在视觉上具有较好的质量,物体具有更加完整的形状和轮廓,并且在KITTI(Karlsruhe Institute of Technology and Toyota Technological Institute)数据集上提升了3D目标检测精度。在使用该数据增强方法后,KITTI数据集下AVOD(aggregate view object detection)检测方法的AP_(3D)-Easy(average precision of 3D object detection on easy setting)提升了8.25%,AVOD-FPN(aggregate view object detection with feature pyramid network)检测方法的AP_(BEV)-Hard(average precision of bird’s eye view on hard setting)提升了7.14%。结论本文提出的激光雷达数据增强算法,实现了点云的稠密化处理,并使3D目标检测结果更加精确。