| 摘要: |
| 针对月面环境下巡视器的障碍物检测问题，通过深度相机获取月面视觉信息和深度数据，以深度神经网络和点云相结合的处理方式实现了对月面障碍物感知功能.针对月面视觉信息，在数据预处理阶段采用多通道数据归一、缩放、镜像及旋转等方法对采集图片进行数据增强，以突出月面特征，并通过single shot multibox detector （SSD）算法进行特征提取和目标定位，实现了对视觉信息的障碍物检测.对物理尺度较大的坡度地形，结合深度信息，通过降采样减轻稠密数据处理的压力，通过主成分分析（PCA）方法计算各点的法向量从而获得坡度的特征信息，实现了准确的月面障碍物感知功能. |
| 关键词: 深度相机 深度学习 点云 月面 障碍物感知 |
| DOI：10.20192/j.cnki.JSHNU（NS）.2025.02.008 |
| 分类号:TP18 |
| 基金项目:上海市自然科学基金（20ZR1440500） |
|
| An obstacle perception method based on depth camera data in lunar surface environment |
|
HUANG Xincheng, GUAN Xiqiang, LIU Yingyuan
|
|
Collage of Information, Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 201418, China
|
| Abstract: |
| To address the issue of obstacle detection in lunar surface environment， visual information and depth data of lunar surface were obtained through a depth camera. The lunar obstacle perception was achieved by combining deep neural networks and point cloud data processing. For the lunar visual information， methods such as multi-channel data normalization， scaling， mirroring and rotation were adopted during data preprocessing stage to enhance the collected images and highlight lunar features. Besides， the single shot multibox detector （SSD） algorithm was employed for feature extraction and object localization， achieving obstacle detection in the visual information. For large physical slope terrains， depth information was combined to reduce the processing pressure of dense data by downsampling. The principal component analysis（PCA） method was used to calculate the normal vector of each point， thereby obtaining the slope feature information. Based on the above method， accurate lunar surface obstacle detection functionality was achieved. |
| Key words: depth camera deep learning point cloud lunar surface obstacle perception |
