视觉SLAM(simultaneous localization and mapping)是智能车辆领域的研究热点,在包含运动目标干扰或近景特征不显著的场景中,容易产生帧间位姿估计结果精度不足或失效问题.为此,本文提出一种结合场景语义信息和路面结构化特征的SLAM算法...视觉SLAM(simultaneous localization and mapping)是智能车辆领域的研究热点,在包含运动目标干扰或近景特征不显著的场景中,容易产生帧间位姿估计结果精度不足或失效问题.为此,本文提出一种结合场景语义信息和路面结构化特征的SLAM算法.首先,针对上述特殊场景中运动目标干扰的情况,设计带有改进金字塔池化模块的语义分割神经网络,得到图像中各像素对应的目标类别,作为剔除运动像素点的依据,从而避免运动点参与特征匹配导致的位姿计算准确性下降问题;然后,针对有效近景特征点不足的情况,基于V视差算法确定图像中的道路平面区域并拟合出精确的视差方程,以计算路面上像素点的精确视差值,并提出一种基于路面结构化特征(车道线、马路边界、路面交通标记等)的位姿计算方法;最后通过场景实验得出,本文提出的改进算法计算结果的绝对轨迹误差小于原算法.证明该方法能够在存在运动目标干扰或缺乏近景特征的场景中具有较高的位姿估计精度,建立了有效的包含语义信息的稠密点云地图,具有良好的环境适应性.展开更多
Core-shell MoSi_(2)@Nb powder was prepared by electrostatic layer self-assembly method.The surfactants SDS(CHSO4Na)and CTAB(C19H42BrN)were used to modify the surface of the two particles to make them charged,and the Z...Core-shell MoSi_(2)@Nb powder was prepared by electrostatic layer self-assembly method.The surfactants SDS(CHSO4Na)and CTAB(C19H42BrN)were used to modify the surface of the two particles to make them charged,and the Zeta potential of the suspension was tested by the Zeta potentiometer.Scanning electron microscope,transmission electron microscope and energy dispersive spectrometer were used to characterize the phase,morphology,microstructure and element distribution of synthetic materials.The results show that when the SDS concentration is 2 mmol/L,the CTAB concentration is 3mmol/L and the pH value of Nb suspension is 5,the coating effect is better after secondary cladding.NbSi_(2)phase is found at the interface between Nb and MoSi_(2)after calcination at 200℃ for 2 h in Ar atmosphere,indicating that Nb is highly active and reacts with Si.Core-shell structure is still retained in MoSi_(2)@Nb material after spark plasma sintering at 1450℃ for 2 h under uniaxial pressure of 40 MPa.However,it is found that Nb reacts strongly with MoSi_(2),and most of the Nb phase is reacted.This issue needs to be addressed in subsequent studies.The fracture toughness(K_(IC))of MoSi_(2)@Nb material is significantly improved to 5.75 MPa·m^(0.5)compared with that of MoSi_(2)material(3.32 MPa·m^(0.5)).展开更多
文摘视觉SLAM(simultaneous localization and mapping)是智能车辆领域的研究热点,在包含运动目标干扰或近景特征不显著的场景中,容易产生帧间位姿估计结果精度不足或失效问题.为此,本文提出一种结合场景语义信息和路面结构化特征的SLAM算法.首先,针对上述特殊场景中运动目标干扰的情况,设计带有改进金字塔池化模块的语义分割神经网络,得到图像中各像素对应的目标类别,作为剔除运动像素点的依据,从而避免运动点参与特征匹配导致的位姿计算准确性下降问题;然后,针对有效近景特征点不足的情况,基于V视差算法确定图像中的道路平面区域并拟合出精确的视差方程,以计算路面上像素点的精确视差值,并提出一种基于路面结构化特征(车道线、马路边界、路面交通标记等)的位姿计算方法;最后通过场景实验得出,本文提出的改进算法计算结果的绝对轨迹误差小于原算法.证明该方法能够在存在运动目标干扰或缺乏近景特征的场景中具有较高的位姿估计精度,建立了有效的包含语义信息的稠密点云地图,具有良好的环境适应性.
基金This research is supported by China’s National Office for Philosophy and Social Sciences(Grant No.:18BYY032)Yangtze University’s College Student Innovation Training Project(Grant No.:Yz2021258).
基金National Natural Science Foundation of China(51861026)The Graduate Innovation Special Fund of Nanchang Hangkong University(YC2022-011)。
文摘Core-shell MoSi_(2)@Nb powder was prepared by electrostatic layer self-assembly method.The surfactants SDS(CHSO4Na)and CTAB(C19H42BrN)were used to modify the surface of the two particles to make them charged,and the Zeta potential of the suspension was tested by the Zeta potentiometer.Scanning electron microscope,transmission electron microscope and energy dispersive spectrometer were used to characterize the phase,morphology,microstructure and element distribution of synthetic materials.The results show that when the SDS concentration is 2 mmol/L,the CTAB concentration is 3mmol/L and the pH value of Nb suspension is 5,the coating effect is better after secondary cladding.NbSi_(2)phase is found at the interface between Nb and MoSi_(2)after calcination at 200℃ for 2 h in Ar atmosphere,indicating that Nb is highly active and reacts with Si.Core-shell structure is still retained in MoSi_(2)@Nb material after spark plasma sintering at 1450℃ for 2 h under uniaxial pressure of 40 MPa.However,it is found that Nb reacts strongly with MoSi_(2),and most of the Nb phase is reacted.This issue needs to be addressed in subsequent studies.The fracture toughness(K_(IC))of MoSi_(2)@Nb material is significantly improved to 5.75 MPa·m^(0.5)compared with that of MoSi_(2)material(3.32 MPa·m^(0.5)).
